At a recent industry conference, Marianne Lake, JPMorgan Chase’s CEO of Consumer and Community Banking, commented on Chase’s branch buildout:

“Increasing branches is directly correlated with deposit growth, with new branches accounting for 40% of JPMorgan Chase's new deposit share gains.”

Other banks have been paying attention. Bank of America, PNC, and Truist have all announced branch network expansion plans, reversing years of closures.

According to Lake:

"We don't expect not to compete. So I would never underestimate the competition. They will build branches. They will be successful. We just have to do it better."

My take: Somewhere between Chase’s 2025 Investor Day presentation and the financial press, Chase’s branch story has been twisted—with important implications for how other bankers should think about the role of branches.

What Chase Actually Said

At the 2025 Investor Day, Lake presented data showing that new branch builds have contributed approximately 40 basis points of retail deposit share gains since 2019.

Chase’s retail deposit share at the time was 11.3%. That means branches opened between 2019 and 2025 accounted for roughly a third of the 120-basis-point total share gain Chase achieved over that period.

One publication, however, reported from the same presentation: “Lake said that new branches built in expansion markets are already responsible for 40% of the bank’s total retail deposits.”

That’s not correct. Forty basis points of market share gains and 40% of total retail deposits aren’t in the same neighborhood. The former is a contributing factor in a multi-pronged growth strategy. The latter would be one of the most remarkable branch-driven deposit stories in American banking history.

The Retail Digital Reality

Another important data point: in 2021, Chase reported that customers opened approximately 50% of consumer deposit accounts digitally.

If 50% of accounts were opened digitally in 2021, the 2026 percentage is likely higher. Chase hasn’t disclosed a more recent number, but hasn’t suggested the trend reversed.

So what? The branches Chase is opening aren’t primarily winning retail consumers. That battle has largely moved online.

According to my research, fintechs' share of new checking/payments accounts grew from 49% in 2024 to 56% in 2025. The megabanks'—BofA, Chase, and Wells Fargo—2025 share of new accounts opened was about half of what it was in 2020. Year over year, Chase's share fell from 5.2% to 3.7%.

Lakes’ branch story—break-even in under 4 years and “meaningful contributions” to deposit gains—is genuine. But “meaningful contribution” does a lot of work in that sentence when retail consumer acquisition has been trending toward digital for years.

The Small Business Story Is the Real Story

The branch narrative makes more sense when you look at the business banking side.

The 2025 Investor Day data showed Chase’s Business Banking primary share at 9.7%—up 20 basis points year-over-year and up 170 basis points since 2019. That trajectory is steeper than the retail deposit share gain, and it’s the kind of growth that branches are well-suited to drive. The question is why are they ‘well-suited.”

New Chase branches don’t just put up a sign. They:

1. Deploy Business Relationship Managers—bankers assigned to small business portfolios—in local markets;

2. Create presence with centers of influence—accountants, attorneys, commercial real estate brokers—who refer business owners; and

3. Give wealth advisors a home base to serve affluent households the bank has identified as holding more than $4 trillion in investable assets outside of Chase.

The physical footprint is the basis for a relationship model that digital hasn’t replicated at the small business level (yet).

The 2025 Investor Day data showed that Chase increased the number of small businesses covered per Business Relationship Manager by 60% since 2019.

That’s the real operational story behind the branch expansion—not foot traffic from retail consumers who increasingly open accounts on their phones.

Chase isn’t exactly starting from behind. According to a new Cornerstone Advisors survey, 20% of small businesses ($100k to $100m in revenue) have a banking or borrowing relationship with Chase—second only to Bank of America at 24%.

Conclusion: Chase’s branch buildout is a long-term push to close that gap and extend its lead over everyone else.

The Fintech Problem Nobody Is Talking About

The logic of Chase’s small business branch strategy rests on an assumption: winning small business banking relationships requires physical presence. It’s an assumption that a growing set of fintechs is invalidating.

Fintechs like Brex, Ramp, or Mercury have built traction by providing a great small business banking digital experience—cash flow visibility, payments, integrated tools, and, increasingly, credit. A new Cornerstone Advisors survey shows that 12% of small businesses now have a relationship with a fintech.

The “fintechs are eating small business banking” narrative looks premature when you put actual market share numbers on it. But that’s exactly what someone would have said about consumer checking accounts in 2019, before Chime scaled to tens of millions of customers.

The fintech aren’t trying to be Chase—they’re trying to make Chase’s Business Relationship Manager model obsolete.

The question isn’t whether small businesses prefer a relationship banker today. Many do. The question is whether the branches Chase continues to open will be the right asset when small business owners are looking for a banking relationship in 2030.

What This Means for Everyone Else

Banks rushing to announce branch expansion programs are taking the wrong lesson from Chase’s success—and possibly at exactly the wrong moment.

Branches aren’t doing what the headlines imply. Retail deposit acquisition has shifted to digital. Chase’s branch buildout is for business relationships, wealth management deepening, and physical presence in markets the bank previously didn’t serve. Those are legitimate strategies with legitimate ROI—for now.

Chase has the scale, the capital, and the right strategy (for now) to plow a ton of money into new branch development. The big question: Do the banks imitating them have that too?

Come into the Fintech Snark Tank and tell me why I’m wrong.

Community banks and credit unions are committed to improving their customers’ and members’ financial health. Whether what they’re doing works is another question.

The prevailing implicit formula—financial health = education + literacy—is a myth. Consumers don't want financial education—they want tools.

The research I’ve done shows that people will pay for those tools: nearly 4 in 10 Gen Zers want credit building, bill management, and subscription management tools bundled with their checking account, and are willing to pay for them.

I’ll stand by my assertion that tools > education. But that’s overly simplistic. Different consumers need different tools—which requires a segmentation of consumers by financial health/wellness drivers and needs.

A new white paper presents an approach that banks and credit unions can use. The report builds a two-dimensional framework that combines short-term financial control with long-term financial security. The result is six consumer segments:

• Established. Highest financial health tier. Strong day-to-day liquidity management and robust long-term capital planning. Heavily skewed toward postgraduate education and high income (73% earn above $300K annually).

• Resilient. Moderate short-term control and adequate long-term security, but with suboptimal financial habits. Think solidly middle class but without optimization of financial results/performance.

• Short-Sighted. Adequate liquidity and can cover current expenses, but little to no long-term planning. Income is directed at immediate consumption. Constrained by education/financial literacy gaps—only 29% hold a bachelor’s degree or higher.

• Illiquid Planners. Have built long-term assets and can plan for the future, but struggle to cover day-to-day demands. Their short-term weakness isn’t behavioral—they’ve been hit by a shock (e.g., job loss or business failure).

• Precarious. Theoretically solvent but practically vulnerable to any income disruption. Latent instability with limited control over their finances. One bad event away from sliding into the bottom tier.

• Distressed. Immediate liquidity crises and minimal long-term safety net—80% lack an undergraduate degree and 66% earn under $50K.

Here are three key findings from the research financial institutions should focus on:

1) Mindset Beats Metrics

A key finding is what the author calls subjective dominance—how an individual’s self-perception of their financial situation outranks their objective financial behaviors and their financial literacy as a predictor of which segment they fall into.

Translation: How people feel about their finances matters more than what they actually do with their money or what they know about it when segmenting them.

Why this is important: Most financial wellness programs are built on the assumption that information and behavior change are the primary levers—i.e., give people better data about their spending, teach them about compound interest, show them their credit score trajectory, etc.—to fix the literacy gap and the behavior follows.

The research suggests that framing is backwards. If perception is the structural driver, then someone who feels financially out of control will likely remain behaviorally out of control, even after completing a financial literacy module or downloading a budgeting tool.

The psychological dimension—the mindset—isn’t a downstream outcome of better financial behavior. It’s an upstream constraint on it.

A financial wellness program built around this insight would include a combination of progress signaling, positive reinforcement loops, and human touch points that change how consumers narrate their own financial situation.

That’s not easy to do or to build, but it beats a scorecard.

2) Not All Strugglers Struggle the Same Way

The report distinguishes between two vulnerable or struggling segments: the Short-Sighted and the Illiquid Planners.

Both look similar on the surface—financially stressed, not fully in control, candidates for financial wellness intervention. But the underlying mechanics are different.

The Short-Sighted have adequate short-term liquidity but weak long-term planning. Their shortcoming is educational: only 29% hold a bachelor’s degree or higher and their financial literacy scores are low.

They don’t have the cognitive tools to think about the future in financial terms. If you’re going to intervene with this group, literacy programs and structured savings nudges are directionally right.

Illiquid Planners, however, have solid long-term assets but not short-term liquidity. Their problem isn’t ignorance—they know how to plan. They’ve been hit by an external shock—job loss, business failure, medical expenses—and their long-term asset base doesn’t help them cover the bills. Their subjective perception scores drop accordingly, which paralyzes their decision-making.

Illiquid Planners don’t need a lesson on compound interest—they need emergency liquidity access, income smoothing products, and an institution that can help them avoid liquidating long-term assets at the worst possible moment.

How many banks or credit unions have built products—e.g., high-yield liquid emergency accounts, short-term income bridge loans, automatic sweep mechanisms that protect long-term savings when short-term cash flows go negative—for that segment? I’m betting the answer is “very few.”

Sending the same financial literacy content to both groups—which is what most wellness programs do—helps one group modestly and does nothing for the other.

3) Wellness Programs Serve the Wrong Segments

Combined, the Established and the Resilient are consumers who are already engaged with their finances, already saving, already thinking about the future. They’re also the members most likely to log into a financial wellness portal, complete a financial health assessment, and engage with educational content.

The Distressed and the Precarious—the consumers who need intervention the most—are the least likely to engage with a self-directed financial wellness tool. Their problem isn’t access to information. Their problem is an accumulation of structural disadvantages exacerbated by income shocks and negative self-perception.

Financial Health Programs Need an Overhaul

Most programs optimize for engagement metrics that naturally favor already-engaged consumers, and then report those metrics as evidence of impact.

Consumers who actually move up the hierarchy rarely do so because of financial wellness content. They do so because of a product, a loan, a refinancing, a counselor conversation, or a life circumstance change.

If a financial health strategy produces strong engagement numbers among your already-healthy members while financially distressed members remain distressed, you don’t have a financial wellness program.

Bottom line: the one-size-fits-all financial wellness program is the wrong model, not just in theory but in demonstrated empirical fact. Different segments require different interventions. A wellness program that doesn’t address how members feel about their finances is incomplete.

A mobile app with a spending analyzer and a set of financial tips isn’t a solution to that problem. It’s a product that makes your program look active while leaving the most vulnerable consumers exactly where they were.

Come into the Fintech Snark Tank and tell me why I’m wrong.

The Wall Street Journal reported that JPMorgan, Bank of America, Citi, Wells Fargo, and a handful of other megabanks plan to launch a shared tokenized deposit network through The Clearing House, targeting a first-half 2027 launch.

The headline called it “a big move for banks.” I beg to differ.

The most honest quote in the article came from Bank of America’s head of global payments solutions: clients aren’t “beating down the door” for tokenized deposits, but there’s been “some interest” and the network will ensure banks are well-positioned.

My take: The megabanks are building a multi-bank blockchain infrastructure in response to “modest” demand, they haven’t chosen a vendor yet, and they haven’t settled on a name—some call it “the bridge,” others “the chain”— but they’re ready to announce it. That’s not a product launch. That’s a press release about a press release.

What This Is

The tokenized deposit network is, in essence, a defensive infrastructure project.

The megabanks that co-own The Clearing House see stablecoin legislation moving through Washington, see crypto firms pushing deeper into institutional payments, and have decided they need a response that keeps deposits “inside the regulated banking system.”

You mean, inside their banks.

That framing is legitimate and the threat is real. Stablecoins designed for treasury operations and cross-border payments are a direct competitive challenge to banks.

But notice what the network is designed to do: programmable treasury operations, real-time liquidity management, cross-border payments for large multinational corporations. That’s a wholesale institutional play.

It’s not the deposit displacement threat that most banks should be worried about.

The deposit displacement risk that matters for the broader banking industry isn’t JPMorgan losing a corporate treasury client to a stablecoin. It’s consumers and small to mid-size business customers moving balances to fintech platforms, crypto wallets, and yield-bearing accounts.

That problem doesn’t get solved by a Clearing House tokenized deposit network built for Fortune 500 treasury departments.

The Community Bank Opportunity—And the Network Trying to Solve It

The Clearing House is co-owned by JPMorgan, BofA, Citi, Wells Fargo, and a handful of other large banks. The WSJ article notes that the network “will be available to banks across the US.”

That’s too funny. “Available” doesn’t mean designed for, priced for, or even relevant to community banks and credit unions. There is, however, a network being built for them—and it’s further along than the recent announcement.

The Cari Network is a blockchain platform, anchored to Ethereum, that enables digital assets representing customer deposits at chartered banks, including FDIC insurance. KeyBank, Huntington Bank, First Horizon, M&T Bank, and Old National Bank are founding members, with a pilot rollout planned for Q3 2026 and full availability targeted for Q4.

Cari is led by Gene Ludwig, the 27th Comptroller of the Currency and founder of Promontory Financial Group, the industry leader in regulatory compliance and risk management consulting, and IntraFi, which built a network of ~3,000 bank members.

Unlike the Clearing House network—which still hasn’t chosen a blockchain vendor—Cari has already selected ZKsync’s Prividium infrastructure, a private permissioned blockchain that allows regulators to audit activity while keeping transaction data private from other participants.

That matters because one of the primary reasons community banks have stayed on the sidelines of blockchain experimentation is compliance risk around transaction transparency.

Cari is also thinking beyond basic payment settlement: Smart contracts running against underlying deposit accounts, automated sweep logic, back-office automation. Use cases that go well past “faster cross-border payments for corporates” and into the infrastructure that could differentiate a $5 billion bank from a fintech challenger.

The challenge Cari faces—which Ludwig acknowledged in a conversation I had with him, Cari’s COO Louie Giacomini, and Peter Marton, Cari’s Chief Risk & Compliance Officer—is that community bank awareness of the opportunity is low.

The “my customers aren’t asking for this” objection is real and widespread.

What’s Missing in the Megabanks’ BridgeChain

Three questions the announcement doesn’t answer:

• Which blockchain? You can’t evaluate a tokenized deposit network without knowing which blockchain. This matters for interoperability, governance, and transaction costs. The megabanks announced intention, Cari announced a stack.

• Where’s the demand? The article describes large multinational corporations as making up “a lot of the initial demand.” If they were clamoring for a shared multi-bank network, why would the BofA say they’re “not beating down the door”?

• How will it compete with itself? JPM Coin is live on Base for institutional clients, and Citi Token Services is in production across 40+ markets. Both already have proprietary tokenized deposit infrastructure—why would they sign up for a shared network that doesn't yet exist?

Megabanks Will be Megabanks

The megabanks are doing what megabanks do: announcing shared infrastructure to protect shared territory, using the consortium they already co-own, for the clients they already serve.

I’m not being cynical—it’s a rational competitive response to a real threat from well-capitalized crypto firms operating under a more favorable regulatory environment than anyone expected two years ago.

Lost in the "tokenized deposits versus stablecoins" framing that dominates today's coverage—and that the banks themselves are using to position this announcement—is a more important point: it's not an either/or.

Tokenized deposits address an efficiency challenge: reconciliation. Stablecoins address an effectiveness challenge: payments.

A bank that builds tokenized deposit infrastructure while ignoring stablecoins has modernized its back office but left the payments battlefield to Circle and its successors.

The banks that will be best positioned aren't the ones that pick a side—they're the ones that understand the difference.

Jamie Dimon told a room full of analysts that JPMorgan Chase is “on the lookout” for an acquisition in the $10 to $20 billion range.

He was careful to say that any target would need to integrate cleanly, fit the bank’s culture, and enhance core businesses rather than sit as a separate standalone unit. “It can’t be just a pie-in-the-sky type of thing,” Dimon said.

“Pie in the sky” does a lot of work there. Conventional acquisition targets—wealth management platforms and regional banks—are the kind of low-imagination moves you make when you’re building scale, not when you’re trying to build the future.

The future banking battleground is control of the operating layer where financial decisions get made—not balance sheet scale.

JPMorganChase doesn’t need to buy a deposit base. It needs to buy the future. Here’s my shopping list for you, Jamie…

1. Ramp

Starting as a corporate card, Ramp has evolved into an AI-native financial operating system for businesses—corporate cards, expense management, accounts payable, procurement, treasury—and a growing layer of autonomous AI agents that manage spending decisions without human review.

In October 2025, Ramp made more than 26 million AI-driven decisions involving $10 billion in spend, blocked over 500,000 out-of-policy transactions worth $291 million, swept idle cash into investments, and flagged a $49,000 AI-generated fake invoice.

The company crossed $1 billion in annualized revenue last year, doubled its run rate in twelve months, and is cash flow positive.

At a $40 billion valuation, that might be a bit pricy for your budget, Jamie, but here’s why you should still consider Ramp: what does JPMorgan’s commercial banking business look like in ten years if Ramp becomes the operating layer every business client runs on?

JPMorgan owns the balance sheet. Ramp is trying to own the interface through which businesses make financial decisions. History suggests the interface owner eventually captures the economics.

You don’t want to pass on this, Jamie, and end up watching Ramp IPO at $60 billion and then have to spend the next decade losing commercial clients to them.

2. Figure Technologies

Figure is a blockchain-based home equity loan platform that has expanded into a full lending and private credit infrastructure play—automated underwriting, instant funding, and a proprietary blockchain for financial asset origination and transfer.

Figure does home equity lending in days, not weeks. The cost structure that comes with that speed is what every large mortgage lender has been trying to achieve through incremental technology investment for twenty years without getting there.

Figure’s Provenance Blockchain is real infrastructure—not a pilot. It’s originated and traded billions in loans, and is attracting institutional investors who want the liquidity and transparency that tokenized loan assets provide.

Kinexys proves JPMorgan understands tokenization. Figure proves it can be applied to consumer lending at scale. Buying Figure isn't acquiring blockchain technology—it's acquiring a real-world tokenized asset ecosystem.

This is a two-for-one, Jamie. It gets you a lending platform that can compete with fintech originators on speed and cost, and consumer-side blockchain infrastructure that complements Kinexys. And you won’t have to use “crypto” in the press release.

3. Betterment

The conventional wealth management acquisition rationale is about AUM—more assets under management, more fees.

That’s fine, but it’s not the most interesting part of a Betterment deal. The fintech’s 401(k) administration business—Betterment at Work—serves thousands of small/mid-sized employers with a fully digital, low-cost retirement plan product.

Banks keep talking about serving small businesses holistically. Yet most can’t profitably offer retirement plans to firms with 10 or 20 employees. Betterment solves a distribution problem JPMorgan has never solved well.

JPMorgan’s retirement and defined contribution business is institutionally focused and structurally expensive for smaller employers to access.

In March, you launched the American Dream Initiative, Jamie, committing JPMorgan to lend $80 billion to small businesses over ten years and expanding its small business target client base from 7 million to 10 million.

If you’re going to deepen your small business commitment at that scale, you need a retirement product small employers can afford and administer. Betterment at Work is the cake—the consumer robo-advisor is the icing.

4. Sardine

Sardine offers a fraud and compliance infrastructure platform built for real-time payments—something legacy fraud tools weren’t designed for. Legacy tools were built for ACH—batched, delayed, and reversible.

Sardine’s platform was architected for a real-time payments environment:behavioral biometrics, device intelligence, and risk scoring designed for real-time decisioning.

Sardine sits at the intersection of fraud prevention and compliance—functions that large banks typically run in separate silos with separate technology stacks. The fintech’s platform unifies them, which is increasingly important as regulators hold banks responsible for fraud losses on real-time payment rails.

Jamie, your bank processes more payments than nearly every other bank on the planet. Your fraud losses run into the billions. Sardine gives you a fraud infrastructure you could offer as a service to commercial clients who face the same exposure.

Every bank wants real-time payments. Few have real-time fraud capabilities. Sardine could become for fraud decisioning what Visa became for card authorization: infrastructure every institution depends on.

Cost-wise, Jamie, a Sardine acquisition is a rounding error. The deal won’t make headlines but it will produce a material reduction in your bank’s fraud losses—and deepen your relationship with commercial clients.

5. Mirakl

Mirakl is an enterprise marketplace platform: infrastructure that allows retailers (Best Buy, Toyota, Nordstrom) and other B2B companies to operate third-party marketplace models. Billions in gross merchandise value annually.

With its partnership with Mirakl (announced in March), the bank has already signaled that the future of B2B commerce is agentic—AI agents procuring, approving, and settling transactions on behalf of firms without a human in the loop. Acquiring the company isn’t an eCommerce bet—it’s an agentic payments infrastructure bet.

As AI agents become the buyers, marketplaces become the places where those agents transact. Owning Mirakl would put JPMorgan at the center of agentic commerce rather than merely processing payments after the fact. And every enterprise marketplace transaction that flows through Mirakl is a potential JPMorgan payment.

Bottom line: JPMorgan doesn’t need more of what it already has. It needs the operating layer that manages the next generation of financial activity—AI agents, real-time fraud infrastructure, intelligence platforms, and workflow software.

Dimon has said JPMC has to compete at the level of tech firms that have “badly beat” it in the past. Spending $20 billion on a wealth platform or bank won’t do that.

The next decade of banking will be won by the institution that owns the software layer where financial decisions get made. The companies on this list aren't financial products. They're control points.

Come into the Fintech Snark Tank and tell me why I’m wrong.

SoFi launched SoFiUSD, a fully reserved US dollar stablecoin running on Ethereum and Solana, redeemable 1:1 for US dollars. It’s available to SoFi members to buy, sell, hold, and convert inside the app. Press and blog headlines were predictable:

• "SoFi Launches SoFiUSD for 14.7 Million Users as First Bank-Issued Stablecoin"

• "SoFiUSD Goes Live for 14.7 Million Users—How America's First Bank-Issued Stablecoin Is Rewriting the Rules of Digital Payments"

• "SoFi Rolls Out SoFiUSD Stablecoin to Banking App Users"

My take: The “first bank-issued stablecoin” and “14.7 million customers” are the wrong headlines. The consumer app launch is the demo. Galileo is the product.

Partnership: Galileo + Mastercard + BitGo

When SoFi announced SoFiUSD in December, its press release was clear:

“SoFiUSD will enable SoFi to serve as a stablecoin infrastructure provider for banks, fintechs, and enterprise platforms.”

Galileo is SoFi’s payments infrastructure arm. It powers the backend for hundreds of fintech clients—card issuing, processing, program management—who can now plug into 24/7 stablecoin settlement without building blockchain infrastructure themselves.

SoFi Bank will use SoFiUSD to settle its own credit and debit transactions through Mastercard, while Galileo will provide fintechs with stablecoin settlement options via the card network.

Another part of the partnership stack is BitGo, which will provide the underlying stablecoin infrastructure—smart contracts, custody, minting and burning operations, and compliance workflows—through its Stablecoin-as-a-Service platform. SoFi Bank and BitGo Bank & Trust are both OCC-regulated, giving them shared regulatory oversight that most stablecoin partnerships lack.

Overall, the back-end infrastructure enhances the speed, reduces cost, and provides 24/7 settlement. It’s a plumbing play, invisible to end users.

Adoption: Who Will Sign Up for SoFiUSD

The value proposition is real, but it’s not for every Galileo client.

For fintechs running high-volume, time-sensitive, or cross-border payments, stablecoin settlement solves some big problems. Card settlement to merchants typically takes one to three business days and ACH’s next-day or same-day options don’t extend cross-border.

Stablecoin settlement takes less than five seconds, 24/7. For platforms that provide gig economy disbursements, cross-border remittances, or B2B payouts, that speed and availability translate into better liquidity and lower friction. At scale, the fee savings on high-volume transactions adds up.

The other pain point is cutoff times. Galileo clients whose customers expect money to move on a Saturday night have a structural problem that stablecoin rails solve without requiring a FedNow integration or a custom technical build.

Economics: Where SoFiUSD Revenue Will Come From

The SoFiUSD economic play is threefold:

• The shorter-term opportunity: reserve income. Similar to Circle’s model, SoFi will hold dollars backing SoFiUSD tokens in circulation and earn yield on those reserves. SoFiUSD has already reached roughly $100 million in circulation—a rounding error in a $320 billion market, but reserve income scales with B2B adoption, not retail wallets.

• The medium-term opportunity: settlement infrastructure fees. Galileo clients that opt into SoFiUSD settlement will generate transaction volume SoFi can monetize. The fee structure hasn’t been disclosed, but this is capital-light revenue that leverages existing SoFi relationships.

• The longer-term opportunity: enterprise treasury. If SoFi lands meaningful enterprise treasury clients on SoFiUSD rails, reserve income and settlement fees scale well beyond what the Galileo fintech client base alone can generate. It also puts SoFi in a league with a very different set of competitors.

Failure Scenario: How SoFi Could Screw This Up

The infrastructure, partnerships, and regulatory framework—thanks to the GENIUS Act—are in place. Which means the ways SoFi fails here are self-inflicted by:

• Chasing the wrong clients. If SoFi pitches SoFiUSD as a generic efficiency upgrade to every Galileo client regardless of fit, it’ll fall flat. SoFi needs to find the fintechs with cross-border pain, the platforms with cutoff time problems, and the B2B disbursement platforms bleeding money on wire fees.

• Selling efficiency. If SoFi leads with “here’s a cheaper, faster rail,” it’s in a feature comparison against FedNow, RTP, and every other payments modernization initiative competing for the IT budget. The stronger argument is existential: stablecoins are going to capture payments and settlement economics whether Galileo clients participate or not.

• Underestimating the compliance bottleneck. The deals most likely to stall won’t stall on economics. They’ll stall when the proposal hits a compliance officer who starts asking about smart contract risk, custody and key management, governance and upgrade controls, and incident response.

• Overlooking the enterprise layer. The business banking platform pitch—settle transactions around the clock using fiat or stablecoins—is the highest-value segment and the easiest to deprioritize when smaller, faster Galileo client wins are available. Treasury teams move slowly, but a handful of meaningful enterprise clients on SoFiUSD settlement rails is what turns this into a big business.

This Could Be The Start of Something Big

SoFi CEO Anthony Noto has built SoFi into a platform company that earns on multiple layers. SoFiUSD is the next layer. Galileo is already rebranding to SoFi Tech Solutions, a signal that SoFi intends to pull the infrastructure business closer to the parent brand as its strategic importance grows.

When it comes to SoFiUSD, the ~15 million retail members are the proof of concept. The 128 million Galileo accounts are the business. The enterprise treasury layer is where it becomes a huge one.

Come into the Fintech Snark Tank and tell me why I’m wrong.

The Wall Street Journal ran a piece about executives creating AI “digital twins”—trained on their emails, speeches, and interviews—to answer employee questions, deliver conference presentations, and generally pretend to be them at scale.

LinkedIn founder Reid Hoffman’s twin has spoken at 75 events. One HR exec’s AI clone chatted with 3,300 employees. Another executive’s digital double helped a colleague inflate her performance review self-assessment from “nice” to “great.”

My take: This is one of the dumbest ideas to emerge from the current AI gold rush. And there a lot of contenders for that honor roll.

Digital Twins Don’t Live Up to the Promise

The WSJ article purports to show “evidence” for the transformative power of digital twins. The reality is just the opposite:

• Reid Hoffman’s twin occasionally makes things up. It delivers punchlines with the expression of someone reading a terms-of-service agreement. When asked a question about its creator’s favorite ice cream, it went with vanilla (the real answer: mint chip). Hoffman shrugged it off as a “data gap that’s improving.”

• Upwork created an AI clone for Kelly Monahan—former managing director of its research institute—that greeted attendees at a conference where she was speaking. Or it was supposed to greet them: Digital Kelly kept stuttering, repeating the same line. Monahan shut it down and told attendees: “This is why we keep humans in the loop. AI isn’t ready for the main stage.” Attendees chuckled.

My take: These aren’t endearing quirks or minor bugs to be patched in the next release. They’re characteristics that: 1) disqualify a system from being used to represent executives, and 2) are symptoms of a mismatch between what LLMs do and what executive communication requires.

The Core Fallacy: Execs Aren’t Bottlenecked by Volume

The entire premise of an AI clone rests on an misconception. The article frames the digital twin as a productivity multiplier because executives don’t have enough time to respond to everyone who wants access to them. Reid Hoffman claims “50% time savings on the weeks it’s deployed.”

But executive value isn’t constrained by response capacity. It’s constrained by judgment which requires context, accountability, and skin in the game. AI clones have none of these—they just automate the appearance of accessibility without the substance that makes access meaningful.

If employees ask questions that a large language model trained on an exec’s old emails and presentations can answer adequately, those questions didn’t need the exec in the first place. If they need the exec, then what they get is an hallucinating impersonator.

An AI Clone is a Legal Disaster Waiting to Happen

The WSJ treats the question of what happens to a digital twin when an executive leaves a firm as a policy issue. It’s more like a looming litigation nightmare.

Execs are trusted because they’re accountable for what they say. AI clones aren’t accountable for anything. They can’t be held responsible, can’t course-correct in real-time, and can’t be fired by the board.

Who owns the professional identity of a person who spent 20 years building their knowledge and communication style? The company that paid them while they were building it? The individual who built it? Both? Neither?

The article notes that “departing employees will be compensated for leaving their digital twin behind.” What happens when Reid AI, still deployed by Greylock after a hypothetical future departure, says something Reid Hoffman would never say?

The article mentions this risk in passing: “an AI double with a supervisory role could create legal headaches.” Ya think?

Organizational Requirements for Digital Twins

I’m not opposed to AI augmenting executive effectiveness. I’m opposed to it being done poorly and calling it “transformational.” If companies want to pursue this in a way that isn’t organizational malpractice, here’s what needs to be in place:

• Transparency. Employees interacting with an AI twin must know—upfront and unambiguously—that they’re talking to an AI system, not the exec. And they should get to opt-out of talking to the twin. Anything less is deception.

• Defined scope. Deployment of a digital twin needs to be narrowly constrained, with guardrails preventing any discussion of employment decisions, legal questions, financial guidance, or personnel matters. The moment a digital twin starts giving HR advice, you need a lawyer in the room, not a chatbot.

• Human review. Any response that touches career development, performance, compensation, team structure, or business strategy should require human sign-off before it reaches the recipient. The twin drafts; the human sends. Anything else is delegating accountability to a system that cannot be held accountable.

• Governance. The ownership, data-retention, and departure questions need to be answered in employment agreements before the twin is built—not improvised when someone quits. Companies that make this up as they go are building legal exposure faster than they’re building productivity.

• Honest measurement. Hoffman’s “50% time savings” is an assertion, not a finding. What did those 75 presentations cost in AI development, human preparation, and reputational risk from the occasional dead-eyed joke delivery? Show me a controlled study, not a testimonial.

What Companies Should Use Executive AI Clones For

Execs accumulate valuable knowledge and communication patterns that are hard to scale and nearly impossible to transfer. AI can help, but not by impersonating the exec in real-time. The right AI applications of this are:

• A knowledge base. Train an AI system on an exec’s documented frameworks, published thinking, and curated decision examples—then make it available to staff as a research tool, clearly labeled as an AI synthesis of that person’s work. Employees can explore how the exec thinks about a class of problems without the system pretending to be that person answering their specific question.

• Training content. An executive’s pattern of thinking about leadership, strategy, or problem-solving can be converted into structured learning content—with AI doing the synthesis and organization work—that is clearly presented as educational material derived from that person’s experience, not as a live stand-in.

• Preparation, not substitution. The performance review use case from the WSJ article is actually fine—if framed correctly. Using an AI system trained on a leader’s feedback style to help employees prepare for difficult conversations is a legitimate productivity tool. The problem isn’t the application. It’s presenting it as the executive showing up, rather than as a preparation aid.

An exec’s judgment should shape the AI’s output, but the exec remains accountable for what gets communicated and decided. The twin assists, not impersonates.

The Bigger Problem: Mistaking Scale for Value

Digital twins are seductive because they address a real frustration that there isn’t enough of the best people to go around. The approach feels like “technology” but it’s mostly illusion.

Real executive leverage comes from building teams that don’t need constant access to the person at the top. It comes from codifying decision frameworks, creating institutional knowledge, and developing second-line leaders.

A digital twin doesn’t do any of that. It just creates a simulation of leadership accessibility. Employees still need the real person’s judgment on hard calls.

Hoffman and the other execs cited in the WSJ aren’t representative of the broader executive population. The vast majority of execs aren’t looking for AI-generated digital twins to do their conference presentations for them. What most execs need aren’t “clones” but systems with the following capabilities:

• Tool: a reactive system that does only what the exec asks at that moment.

• Assistant: a system that can plan a few steps, use approved tools such as search or calendars, and propose next moves while still deferring decisions to the exec.

• Operator: a system that handles a workflow end-to-end authorized by the exec.

• Collaborator: a system that’s able to initiate and coordinate a multi‑step work flow without constant prompts but operates within guardrails and requires the exec’s approval for certain decisions.

• Autonomous actor: a system that can perceive the environment, make decisions, and take action without the exec’s involvement.

Adoption Isn’t Validation

Reid Hoffman says he expects every company with 50+ employees to have digital twins for senior management within ten years. Maybe. But expecting widespread adoption of a technology doesn’t mean the technology is a good idea.

Think about it: Nobody’s perfect. Why would/should companies spend a lot of money to replicate imperfection?

The personal computer genuinely multiplied individual productivity because it automated tasks that were mechanical—calculation, formatting, document retrieval. The digital twin is trying to automate something different: the human credibility that makes leadership communication mean something.

That’s not a productivity problem. It’s a category error. No amount of multilingual conference appearances is going to fix that.

Come into the Fintech Snark Tank and tell me why I’m wrong.

I asked Claude to check out the websites of a dozen or so agentic AI vendors selling to banks and credit unions in an effort to answer a simple question: what are they telling the market about what their tools do?

Across the board, the vendors led with outcomes: faster decisions, higher throughput, reduced analyst time, more approvals without additional risk.

One vendor’s site boasts about going from “lead to term sheet in minutes, not days” before explaining how its platform works. Another’s homepage stacks three metric callouts above the fold before offering a sentence of product description.

I get the logic. CEOs respond to efficiency, cost, and revenue metrics. Explaining what an AI agent actually does inside a lending workflow is a harder homepage headline than “80% faster decisions.” Vendors optimize for the pitch that closes deals.

The problem: Unlike technologies they’ve purchased before—digital banking platforms, loan origination systems—bankers don’t understand how AI tools get to “80% faster decisions.”

After signing a contract for a generative or agentic AI tool, bank execs have to:

• Address the board’s questions about model governance;

• Explain how credit decisions are made and who’s in the loop to examiners; and

• Make the Chief Compliance Officer comfortable enough to let the tool run.

“80% faster decisions” doesn’t help with any of that.

What banks need—but what not all vendors provide—is a plain-language explanation of the vendor’s AI: what data it uses, where that data comes from, what it does with the data, how it handles exceptions, where humans are in the loop, how it embeds the bank’s credit policies, and more.

That’s not a nice-to-have for banks and credit unions—that’s the purchase decision.

Pharmaceutical sales is a good analogy. Drug companies figured out decades ago that selling to physicians requires more than efficacy data. Physicians need to understand how the drug works before they’ll prescribe it, because they’re the ones who have to explain it to patients and defend it to peers. Pharma reps who lead with “patients recover 40% faster” but can’t explain the underlying biology don’t get far.

Tech vendors selling AI to bankers are in the same position.

What the Pitch Promises vs. What the Examiner Asks

Here’s the gap between what vendors pitch and what examiners ask:

• Pitch: Our AI tools achieve an 80% reduction in total underwriting time. Examiner: What data inputs does the model rely on, and how do you validate that they’re accurate?

• Pitch: We deliver 35% more qualified leads through consistent policy application. Examiner: How does the system document exceptions, and who has authority to override the model’s recommendation?

• Pitch: AI agents complete end-to-end credit workflows with little human effort. Examiner: Where exactly does human review occur, and how do you demonstrate that humans are meaningfully in the loop rather than rubber-stamping outputs?

• Pitch: AI agents can be decision-ready analysis in less than three minutes. Examiner: How do you test for fair lending compliance, and what’s your process for identifying disparate impact in the model’s outputs?

The examiner questions aren’t hypothetical. They’re drawn from the OCC’s model risk management guidance, the CFPB’s fair lending examination procedures, and interagency AI guidance that’s been circulating since 2021.

Examiners have been building their list of AI questions for a few years. Vendors, for the most part, are still building ROI slides.

Vendors Need to Help Banks Develop an “AI Vocabulary”

A few vendors get it. The smart ones build educational infrastructure into their sites by providing guides on how specific workflows operate, product taxonomies organized by what the AI does rather than what the bank saves, and demos that show the decision logic, not just the dashboard. Most of the market hasn’t gotten there yet.

The irony is that the vendors selling agentic AI—the category where the process-related questions are most pressing—are often the most outcome-heavy in their messaging (based on the websites Claude reviewed).

Agentic AI is new to most bank and credit union buyers. These are systems that take actions, make sequential decisions, and operate autonomously inside a compliance-driven workflow.

The bar for “I understand what this thing does before I deploy it” is higher than it’s ever been. And the pitch is more metrics-heavy than ever.

There’s an opportunity in that gap.

My take: Vendors that build a reputation for explaining the technology in language a credit officer can use in a committee meeting will earn trust faster than vendors promising the best ROI.

What’s missing, in many cases, is a “AI vocabulary” to help the bank discuss AI opportunities, risks, and policies (how many times have I said over the past two years, “we need to stop using the term AI!”?). Banks need vendors to give them that vocabulary, not just the headline numbers.

Take the distinction between robotic process automation and agentic AI. It matters, both for what banks can expect the technology to do and for how it’s governed.

But a lot of vendor messaging treats the two as interchangeable or it glosses over the difference entirely with language like “agents that learn, think, and act.” That phrase—and variations of it—is doing a lot of work in the market right now and it’s not doing it honestly. True agentic AI:

• Sets and pursues goals dynamically rather than executing a predefined workflow;

• Decides about what action(s) to take based on context rather than rules;

• Handles novel situations outside its training parameters; and

• Orchestrates multiple tools or sub-agents autonomously.

That’s a high bar. Most of what’s being sold as agentic AI today doesn’t clear it.

What many of these systems do is more like intelligent RPA—document classification, data extraction, rule-based decisioning, straight-through processing triggered by document inputs.

I’m not saying that’s not useful. Automating 60% of consumer loan decisions by routing the easy ones away from human underwriters and enabling them to focus on the more complex cases is solid execution.

But that’s been happening for years. Calling it agentic AI because there’s an LLM handling document classification creates expectations that the technology can’t meet and governance questions that banks aren’t prepared to answer.

The tell is usually in the vendor’s language. Claiming to have “APIs trained on company data that execute predefined workflows” is sophisticated automation. Calling those APIs “agents” is marketing.

Technology vendors have an opportunity—and an obligation—to close the vocabulary gap rather than exploit it. The ones who do will earn something more valuable than a closed deal: a customer who can explain and defend the purchase.

Bottom line: Speed and cost reduction are table stakes. Smart vendors will help bankers answer the questions the board and examiners will ask: How does this work, and how do you know it’s working right?

Come into the Fintech Snark Tank and tell me why I’m wrong.

The Senate Banking Committee voted 15-9 on May 14 to advance the Digital Asset Market Clarity Act—known as the CLARITY Act—to the full Senate floor. Two Democrats crossed the aisle. The White House has a July 4 signing target on the calendar. Polymarket’s odds of passage jumped immediately after the vote.

The bill still has to survive a full Senate vote requiring 60 votes—meaning at least seven Democrats have to flip sides—a reconciliation with the House version passed 294-134 back in July 2025, and a conference process that could reopen every fight that’s already been fought.

Senators Lummis and Moreno warned that missing the May 21 window will likely push the whole thing past the November midterms into the next Congress.

The important story here is what the banking industry's posture in this fight reveals about how they've been thinking about the deposit displacement problem—and how long they've been thinking about it wrong.

But first, let’s make sure we’re on the same page about what the CLARITY Act does.

What the CLARITY Act Does

Strip away the lobbying noise and the bill does four things that matter:

• Addresses the SEC/CFTC turf war. Digital assets that are “sufficiently decentralized” get classified as commodities under CFTC oversight. Assets that look and act like securities stay with the SEC. Don’t get too excited about this, however, because “sufficiently decentralized” is hardly a well-defined concept. For now, think of Bitcoin and Ethereum as “sufficiently decentralized.”

• Establishes federal rules for stablecoin issuance. This includes reserve requirements and consumer protections. This is deeply intertwined with the GENIUS Act (which already passed the full Senate last year), and the interaction between the two bills is one of the things that needs to get resolved in conference.

• Tackles stablecoin rewards. This is what raises the banking industry’s blood pressure. The bill—as advanced—bans passive yield on stablecoin holdings that function like deposit interest, while permitting activity-based rewards tied to transactions, trading volume, or platform use. Think credit card rewards, not savings account interest.

• Bans CBDCs. The full title of the act is “Digital Asset Market Clarity Act and Anti-CBDC Surveillance State Act.” Whether you think that’s a feature or a bug says a lot about you.

The Banks’ Argument: Fighting a Fire That’s Already Burned

The banking associations—ABA, ICBA, BPI—aren’t in favor of most of this. Their argument: stablecoin rewards, even activity-based ones, would drain deposits from the banking system and reduce the capital available for mortgages and business loans.

Standard Chartered estimated that a yield provision could redirect up to $500 billion in deposits toward stablecoin products by 2028.

What the associations didn’t mention, however, is that deposit displacement is already happening.

According to a Cornerstone Advisors research report, $3.15 trillion left megabanks, regional banks, and community-based financial institutions for fintech savings and investment accounts between 2020 and 2025.

The banking associations have spent months lobbying against a “future” scenario which has been playing out for more than five years now.

Deposit Displacement Isn’t Just a Young Person’s Game

The conventional assumption is that deposit displacement is driven by Gen Zers and Millennials. Not so. Gen Xers and Baby Boomers accounted for 65% of the deposit outflow to fintech investment accounts between 2020 and 2025.

The average boomer moving money to a fintech investment account moves $1,027 per transaction—more than twice what the average Gen Zer moves. Boomers do it less frequently than younger cohorts, but they’re moving larger sums each time.

Community banks and credit unions have framed the fintech competition problem as a battle for young consumers. The data says it’s also a battle for wealth-accumulating older consumers who are quietly reallocating assets to non-bank platforms.

For credit unions, the picture is particularly stark. Nearly half of zillennial (Gen Z + Millennial) credit union members plan to move a banking relationship to a fintech in the next 12 months. Credit unions have historically counted member loyalty as a structural advantage. Among younger members, that advantage is now questionable.

The Paycheck Motel Problem

At the root of the problem is this fact: Checking accounts have become “paycheck motels”—temporary places for people’s money to stay before it moves on the bigger and better places.

Fintechs' share of new checking/payments accounts grew from 49% in 2024 to 56% in 2025. Paypal and Venmo together accounted for about a quarter of all new accounts opened. Add Chime and you're at 37% of the new accounts.

And what do zillennials do with their fintech brokerage accounts once they open them? Nearly 6 in 10 have a checking account at their fintech brokerage. Half have a savings account there. A third have a credit card. The investing account isn’t a satellite relationship—it’s becoming the primary one.

Crypto: Already Embedded in Zillennial Portfolios

Whatever happens in the Senate, the crypto question is already a present-tense issue for financial institutions—not a future hypothetical.

A quarter of Gen Z investors and a third of millennial investors already have crypto in their portfolios. On average, zillennial investors hold about 25% of their investable assets in crypto. One in five has more than half their investment portfolio in crypto. A third of Gen Z and millennial investors plan to invest in crypto in the next 12 months.

The last time zillennials opened an investment account, 45% opened it at a fintech, 17% opened it at a cryptocurrency exchange, and just 1 in 5 opened it at a traditional financial institution.

The framing of “what should banks do if crypto gets regulated” is the wrong frame. The right frame is: what should banks do about the customers who are already investing in crypto through non-bank platforms?

CLARITY Act: Crypto Providers Get Legitimacy, Not Victory

The crypto industry is getting what it says it’s always wanted: a regulatory framework. Coinbase, Circle, Ripple, and Andreessen Horowitz all backed the bill. Coinbase CEO Brian Armstrong reversed his January opposition on May 1 with a three-word post: “Mark it up.”

Regulatory clarity is the foundation everything else is built on. Without defined jurisdictional lines between the SEC and CFTC, institutional capital stays on the sidelines, compliance infrastructure can’t be built with confidence, and every product launch carries litigation risk. DeFi developer protections survived the markup, which was a red line for the crypto industry.

The Path Forward is Narrower Than the Prediction Markets Suggest

The odds of passage on Polymarket jumped from under 50% in late April to the mid- to upper-60%s after the committee vote. I really can’t imagine why considering:

• The ethics fight is unresolved. Several Senate Democrats said they won’t vote yes without language that prevents elected officials from profiting from crypto.

• The AML provisions are contested. Law enforcement groups argue the bill weakens their ability to track illicit financial activity through digital assets—a concern with bipartisan resonance that isn’t going away.

• The legislative path is complicated. The Senate Banking Committee version needs to be merged with the parallel Senate Agriculture Committee bill. The combined bill then needs 60 votes on the floor. Then it goes to conference with the House version. Then it comes back for another vote in each chamber.

Senators Lummis and Moreno warned that missing the Memorial Day window on May 21 likely pushes the whole bill past the November midterms into the next Congress. Whether that warning is negotiating posture or genuine risk assessment isn’t something I can guess about.

In the best case, enforceable rules won’t exist until 2027. What do the people betting on Polymarket know that I’m missing?

Banks Need to STOP Waiting

The instinct of most community banks and credit unions is to wait. Wait for the bill to pass, wait for the rules, wait for the guidance. That instinct is understandable—but wrong—and the deposit displacement data makes clear why.

The $3 trillion that has already left for fintech investment and savings accounts didn’t leave because of regulatory loopholes or ambiguity. It left because fintech platforms offered something banks didn’t: investing, directly integrated with the account.

Among current zillennial investors who would consider investing with their bank or credit union, the top draw was investing directly from the checking account. An integrated digital investing app was a close second. Educational resources and lower fees rounded out the list.

The opportunity isn’t hypothetical. Among non-investing zillennials—the other half—nearly half cited “I don’t know enough about it” and half cited “I don’t have enough money to invest” as their reasons for not investing.

Both of those are solvable problems for a financial institution with a relationship and a digital channel. Seven in ten non-investing zillennials already have more than $5,000 in savings. The capital is there. The nudge isn’t.

That context reframes what community banks and credit unions should do now:

• Build the investing product that Zillennials want. The strongest single driver of zillennial interest in banking-integrated investing is the ability to invest directly from the checking account—not a separate brokerage account, not a referral to a third party, but an integrated experience that keeps the relationship inside the institution. This is a buildable product today, through partnerships, white-label platforms, or embedded investing technology. The CLARITY Act’s passage or failure doesn’t change the consumer demand.

• Heed the crypto signal, not the product. The 25% portfolio allocation to crypto among zillennial investors is a signal about what those consumers want from an investing relationship: access to a broad range of assets, a digital-native experience, and no friction between holding cash and putting it to work. Banks and credit unions that interpret this as “we need to offer Bitcoin” are solving for the product. Those that interpret it as “we need to rethink the entire investing experience” are solving for the relationship.

The CLARITY Act, if it passes, will create a regulated environment for crypto that makes it easier for financial institutions to offer crypto-adjacent products without litigation risk. That’s an opportunity worth preparing for.

But the more pressing opportunity—the one that doesn’t require waiting for 2027 rule making—is the $2 trillion in deposit outflow to fintech investing platforms that has nothing to do with digital assets.

Banks and credit unions that want to stem the outflow have the consumer research, the product roadmap, and the relationship advantage to do it. What the lobbying campaign in Washington keeps avoiding is whether they have the urgency.

For more on what community banks need to know and do about stablecoins, click here.

OpenAI rolled out a preview of a new personal finance experience in ChatGPT for US Pro subscribers. Powered by Plaid, it lets users connect bank, credit, investment, and other accounts across more than 12,000 institutions.

The dashboard shows spending patterns, subscriptions, upcoming payments, and portfolio performance. Users can ask questions like “Why does it feel like cash is tighter this month?” or “Help me build a plan to buy a house in the next five years.”

The claims are ambitious. Plaid CEO Zach Perret called the deal a step toward “greater financial freedom” for consumers.

OpenAI pointed to over 200 million people who ask ChatGPT financial questions every month as proof of pent-up demand and noted the involvement of Ethan Bloch—founder of the savings app Digit—as the architect of its consumer finance ambitions.

The message is clear: this time, AI-powered personal finance is going to work.

I’m not buying it.

OpenAI’s Personal Finance Acquisitions

The Plaid deal didn’t arrive in a vacuum. In the six months before this launch, OpenAI made two fintech acquisitions including:

• Roi. In October 2025, OpenAI acquired Roi, a personal finance app focused on helping users track and optimize their financial lives.

• Hiro Finance. Acquired in April 2026, Hiro Finance was founded by Ethan Bloch, the entrepreneur who built and sold Digit to Oportun for roughly $230 million.

Hiro barely existed. The 10-employee company—which raised $6.3 million in seed funding from Ribbit, General Catalyst, and Restive—launched its AI personal finance tool about five months ago.

OpenAI shut the product down immediately upon acquisition and deleted all user data. Conclusion: This was a talent acquisition—not a tech buy—a bet on Bloch’s track record, not on anything Hiro had actually built or proven.

It’s worth noting that Digit didn’t succeed by providing advice. It succeeded by moving money automatically, in small amounts calibrated to what users could afford. The insight was behavioral, not analytical: remove the decision, and people save.

ChatGPT’s financial access is currently read-only. Perret signaled the direction, however: “My hunch is that what we launch with is not what we’ll end with, in a year or two or three years.”

Two fintech acquihires, both focused on consumer financial behavior, both pointing to the same destination: AI that doesn’t just give advice but does something. Whether OpenAI gets there is the question.

The Deal is a ChatGPT Pro Upsell/Retention Play

Before accepting the consumer financial empowerment claim, it’s worth looking at the economics here: who’s paying whom, for what, and why?

OpenAI is the customer in this deal. They’ll pay Plaid for data access—a mix of per-connection, subscription, and per-API-call fees—that typically run $0.30 to $1.50 or more per user depending on volume and product tier. As the user base scales (ok, if…) OpenAI’s costs rise—but so does its leverage to negotiate rates.

The more important economic question is why OpenAI is doing this in the first place. Two big drivers:

• Revenue. ChatGPT Pro carries a high monthly price tag—one that requires ongoing justification for subscribers. Usage—or engagement—is the name of the game here. OpenAI’s bet is that once a user links their bank accounts and spends a few months receiving personalized advice from ChatGPT Pro, canceling the subscription will become hard to do.

• Competition. Intuit is a major threat to OpenAI with products like Credit Karma, Mint, and TurboTax which have tens of millions of users that already pay for the services. With integration via a conversational AI interface reportedly coming, the substitution pressure from Intuit’s consumer franchise is real.

For OpenAI, the financial co-pilot is the hook. Revenue growth through engagement is the product. Consumer financial wellness, if it happens, would be a nice byproduct.

Why This Won’t Work the Way OpenAI Claims It Will

OpenAI’s logic: 200 million people ask ChatGPT financial questions every month, therefore there’s enormous latent appetite for personalized, account-linked financial advice that will be worth paying $100 (or more) per month for.

My take: It’s a house of cards. Here’s why:

1) 200 million ChatGPT users aren’t looking for serious financial advice. Many are asking: How do I buy Bitcoin? What’s the best meme stock right now? How can I make passive income? Can I retire on $400k? Should I put my savings in a high-yield account or the S&P 500? Those aren’t users waiting to connect their bank account and build a structured debt payoff strategy. They’re looking for a shortcut to get rich. No amount of Plaid integration converts them into engaged personal finance managers.

2) Plaid doesn’t serve all of the “200 million.” The 200 million is a global number, which means that some chunk of those ChatGPT users are in geographies not served by Plaid. The current addressable market is the subset of those users who have stable, multi-account finances, genuine planning needs, enough trust in OpenAI to share their financial data, and—for now—that live in the US. That’s a much smaller number.

3) The “advice” will be limited. There are structural problems even for the users who do connect. Transaction and balance data is cash flow data, not a financial picture. Home equity—the largest single asset for most American households—won’t show up. Neither will vested RSUs, 529 balances, whole-life cash value, or pension obligations. Advice built on that partial view isn’t personalized financial planning.

4) The advice isn’t actionable. The gap between knowing and doing isn’t a data problem—it’s a friction problem. What the “200 million” get is “advice,” not action. That gap is important. People know they overspend and know they should save more. Without the ability to act on the user’s behalf—to move money, cancel subscriptions—ChatGPT won’t impact financial health. It beats a pie chart, but it’s not Digit.

5) The liability issue. AI has no fiduciary duty or legal obligation to act in a user’s best interest. Without it, ChatGPT’s recommendations are more like those of a confident friend than an advisor. “We said it wasn’t professional advice” has never been a successful defense in a class action suit. The argument, of course, will be “we didn’t charge for advice, we charged for data access.” Some disgruntled user is bound to sue.

Bottom line: This is a good deal for Plaid, it’s a questionable revenue play for OpenAI, and a bad deal for a lot of ChatGPT Pro users who’ll end up paying ~$1,000/year for it.

OpenAI and Plaid can talk all they want about improving financial health and wellness for the “200 million” asking for “financial advice” every month. The financial health impact won’t happen simply because people can link accounts.

And if the advice is any good, it will tell those ChatGPT Pro users that the service isn’t worth the additional monthly cost. Too bad it won’t be able to auto-cancel it for them.

Fiserv announced the launch of agentOS, billing it as “the operating system for agentic AI in banking.” According to the company’s press release:

“Built to operate natively across Fiserv’s platforms—core, payments, issuer processing, and servicing—agentOS enables banks and credit unions to move beyond disconnected agentic pilots to an enterprise-grade deployment with policy controls, auditability and human oversight embedded in the design. agentOS features the industry’s first agent marketplace built natively for banking workflows allowing financial institutions to access Fiserv-built agents, build their own, or deploy third-party agents within a controlled architecture.”

Everyone in banking technology has apparently decided that “agents” is the word that unlocks budget conversations in 2026. The questions worth asking: what’s real, what isn’t, and what should Fiserv’s bank and credit union clients do?

An “Operating System”? Really?

The headline says “Fiserv launches agentOS,” but the more accurate version is “Fiserv announces it will launch agentOS later this summer, with two institutions currently piloting it.” That’s not nothing, but it’s not something you can deploy next month.

More importantly, “operating system” is doing a lot of work in the headline.

An operating system manages hardware resources and provides a common interface for applications across diverse environments. What agentOS actually is—based on the announcement—is a governance and orchestration layer for AI agents that run natively within Fiserv’s existing platforms.

That’s valuable, but it’s not really an operating system. It’s closer to what the enterprise software world would call a middleware or orchestration framework.

This matters: the term “operating system” implies something that’s neutral, universal, and platform agnostic. The product is the opposite of that. agentOS is “built to operate natively across Fiserv’s platforms. If your institution runs Fiserv’s core and payments infrastructure, agentOS extends your investment. If it doesn’t, agentOS is not your operating system. It’s Fiserv’s.

The marketplace is where the lock-in strategy becomes visible. Third-party agents deployed “within a controlled architecture” means those agents run under Fiserv’s governance framework, on Fiserv’s infrastructure. For Fiserv clients, that’s a feature. For institutions evaluating multi-vendor strategies or looking to bring in best-of-breed AI tools from outside Fiserv’s ecosystem, it’s a constraint.

The Agentic AI Competitive Landscape

“First governed agent marketplace built for banking” is a phrase worth interrogating. FIS has been building an agent-first governed environment simultaneously, with Anthropic’s forward-deployed engineers embedded since May 4. Fiserv is in a three-way announcement war with FIS and nCino, with everyone else circling:

• FIS announced its Financial Crimes AI Agent with Anthropic earlier in May (see my take here). FIS’s stated roadmap beyond AML includes credit decisioning, deposit retention, customer onboarding, and fraud prevention. Very much like Fiserv’s play, just ten days earlier and with a different AI partner.

• nCino has gone furthest in repositioning around agents, rebranding itself outright as “the platform for agentic AI banking.” Its role-based agents are built around banking jobs: Executive, Analyst, Service, Processor, and Client.

• Jack Henry’s partnership with boost.ai, announced in June 2025, targets contact center automation. I’d argue that the use of the term “agent” here is more akin to “call center rep” or agent, than to an “AI agent” that perceives the environment, makes decisions, and take actions without human direction.

• The emerging third-party ecosystem is worth watching regardless of which core you’re on. Firms like AgentFlow, EnFi, TransientAI, and Persado are building agentic AI layers that sit on top of existing core infrastructure. Cornerstone’s research found that roughly a third of banks plan to rely on specialty fintech providers (rather than their LOS vendor) for commercial lending agentic AI.

AI Agents in Commercial Lending

Fiserv’s announced initial use case of commercial loan onboarding is solid. Commercial loan onboarding is one of the more painful processes in community banking—manual, multi-system, slow. Daily operational reporting being cut from 10 minutes to seconds, as Boulder Dam’s CEO noted, is a real win.

There’s a real split in perspectives among banks, however, about the use and impact of AI agents—and AI, in general—in commercial lending. Roughly a little more than half of bankers expect AI to have a strong or moderate impact on the various commercial lending processes. But there’s between 20% and 30% who believe it will have a minimal impact, and about 1 in 5 who aren’t sure what the impact will be.

The divergence in perspective is important—it reflects basic differences in:

1. Views about the role of technology in the business. The execs represented on the left of the chart are (generally speaking) see the potential not just for AI impact but for broader technology impact on commercial lending. Those represented by the light gray bar stick to the “this is a relationship business” and “we need a guy in [that town next to the one we’re already in] in order to grow” perspective.

2. How educated about AI bank execs are right now. It’s 2026—senior bank execs shouldn’t “not be sure” what impact AI is going to have on their commercial lending business. I can understand why there are execs who think AI will have “minimal” impact—it’s inexcusable to not have a perspective.

3. How banks execs will receive agentic AI announcements. With roughly a third of bank execs in the “minimal impact” to “not sure what the impact will be” categories, providers of agentic AI “operating systems” and solutions shouldn’t expect a rush of interest and commitment to the new offerings. Kudos to the banks named in the Fiserv and FIS announcements—but for now, they represent a tiny fraction of agentic AI-ready banks.

What Fiserv Clients Should Do

Commercial loan onboarding and AML triage are high-value, well-defined problems. If your institution does significant commercial lending or has a high volume of AML alerts, get into the conversation with Fiserv now about the beta program or summer deployment timeline. Don’t wait for August general availability if your business case is clear. For the rest of the Fiserv client base:

• Get past the “operating system” framing. agentOS is a governance and orchestration framework for Fiserv-native workflows. The relevant questions are: what agents are available for my specific use cases, what are the implementation timelines and costs, what data and integration requirements are expected, and what are the contract terms around agent performance and liability?

• Evaluate your data readiness. As I’m fond of saying, “if your data quality is low, your AI strategy will blow.” The most common reason why AI pilots fail or underperform at is that the underlying data doesn’t support the agent’s decision logic. Before committing to any agentic AI deployment—from Fiserv or anyone else—take a hard look at the quality of the data in the workflows the agent will touch.

• Build your governance infrastructure now. Human-in-the-loop checkpoints, audit trails, model documentation, and decision traceability are required for agentic AI deployment. Fiserv’s governance-by-design approach is a legitimate selling point. But governance doesn’t live in the vendor’s platform—it lives in your policies, procedures, and documentation. That needs to be in place, regardless of the vendor chosen.

• Don’t assume Fiserv’s marketplace is your only option. The agentOS Marketplace is Fiserv-native by design. If your institution has specific use cases in areas like lending automation, fraud decisioning, or deposit analytics, the emerging third-party ecosystem may offer more flexibility, faster deployment, and more competitive pricing.

The Bigger Questions

The AI agent race among core and banking technology vendors is real, and the competitive pressure is healthy. Fiserv’s agentOS, FIS’s Financial Crimes agent, and nCino’s Digital Partners all represent serious investments in infrastructure that community banks and credit unions will eventually rely on. The governance frameworks are more sophisticated than the first generation of fintech AI deployments and the use cases are better targeted.

Announce-now, deliver-later is common in banktech. The “operating system” and “platform for agentic AI banking” positioning from the vendors is a signal that this is as much a competitive marketing moment as a product delivery moment.

For Fiserv’s clients, the question isn’t whether to engage with agentOS—it’s worth engaging with seriously. The question is whether to engage with it exclusively, and whether to engage now or after the August cohort proves out the architecture at scale. Both of those questions deserve more than a vendor briefing to answer.

The big question for all financial institutions is this: Using AI agents, how fast can you get—at what rate, at what cost, and at what operational risk?

Jamie Dimon and Anthropic’s CEO Dario Amodei walked on stage together in New York last week and proceeded to answer a direct question—“Is the freakout over AI-enabled cyberattacks warranted?”—without really answering it.

Which, if you think about it, answered it. Here are the facts:

• Anthropic’s Claude Mythos Preview is a general-purpose frontier model that has turned out to be strikingly capable at computer security tasks. It’s already found thousands of previously unknown software vulnerabilities—vulnerabilities that survived decades of human review and millions of automated security scans.

• Dimon, who had downplayed the cybersecurity risks of advanced AI during JPMorgan’s earnings call, showed up at the Anthropic event and definitively claimed that Mythos is a “very heightened risk.” He urged Amodei to lay out the details and “give people a chance to study it, understand the vulnerabilities, come up with plans so we can handle it.”

• Earlier, Treasury Secretary Bessent and Federal Reserve Chair Powell convened an emergency meeting with Wall Street CEOs to discuss the systemic risks Mythos poses to the US financial system. The Fed’s Vice Chair for Supervision Michelle Bowman summed it up: “This capability enables firms to address self-identified vulnerabilities, thereby enhancing cybersecurity. If used maliciously, it could be deployed to identify and exploit weaknesses.”

Is it time for the banking industry to panic?

Here’s the case for panic:

• Anthropic briefed government officials on Mythos’s offensive and defensive cyber applications, and has been in ongoing discussions with CISA and the Center for AI Standards and Innovation. When an AI company sits in emergency sessions with regulators, you’ve got to pay attention—this isn’t theater.

• Amodei said there’s a 6 to12 twelve month window to patch Mythos’s vulnerabilities before Chinese AI capabilities catch up. The clock on who patches them first—defenders or attackers—is running.

• Dimon pointed out: “Banks are attached to exchanges and all these other things that create other layers of risk.” Connectivity is the problem. A bank may run a clean shop, but it’s still connected to third party vendors who may not be.

Here’s the case for a calmer response:

• Cybersecurity experts claim that the vulnerabilities Mythos is surfacing can be found using existing models. The bad actors who were going to come after your bank were already doing so with the tools they had.

• Anthropic has been warning for months that AI’s cyber capabilities were advancing rapidly, pointing to a blog post showing that Claude Opus 4.6 had already found more than 500 high-severity vulnerabilities in open-source software.

• Dimon said fundamentals still apply: “A lot of it’s hygiene—how do you protect your data, networks, routers, hardware, changing your passcode? Doing all those things right dramatically reduces the risk.” That’s good news for institutions that don’t have the $17 billion tech budget Jamie’s bank has.

The 3 Problems Banks Didn’t Know They Had

The Mythos “freakout” is partly about a model that did what previous models were already doing—just faster and at orders-of-magnitude greater scale. Mythos didn’t create the threat—it just accelerated and surfaced it.

Here’s my take:

1. This is a vendor management problem. Anthropic is disclosing the vulnerabilities directly to the software vendors responsible for patching them which means fixes are in the pipeline. But a fix in the pipeline isn't a fix in production. If your core processor and/or digital banking vendor can’t tell you which vulnerabilities affect your systems and their timeline for patching them you have a vendor management problem.

2. This is a compliance problem. Dimon said: “In the old days, you put out a patch, people had a week or two to fix it. Now you say it’s got to be like minutes.” Most banks and credit unions aren’t patching things that fast—and they might not need to be—but they do need to close the gap between “patch available” and “patch applied” from weeks to days. That’s a process and governance issue, not just a technology issue.

3. This is a board/executive team problem. Cybersecurity risk belongs on the agenda of board and executive team meetings as a strategic conversation—not a checklist exercise—about the bank’s tech security dependencies. Amodei’s 6-12 month timeframe is a window for vendors to close the most critical gaps. The bank’s job is to hold them accountable for doing it—and to decide what to do about it if they don’t.

Dimon is right that Mythos represents heightened risk. He’s also right that the answer is preparation, not paralysis. Banks and credit unions need to treat this as a strategic moment to sharpen their vendor governance and board-level cyber literacy.

So, is the freakout over AI-enabled cyberattacks warranted? The threat is real, but the hysteria is optional.

For decades, the dominant strategic conversation in banking has centered on technology as a source of competitive advantage and differentiation.

Acquire the right core system, deploy the right digital platform, build the right data infrastructure. The story line was that institutions that executed on these imperatives would develop competitive advantages that competitors couldn’t easily close.

That era has ended.

When any bank—not to mention two 14-year-olds in their bedroom—can build and deploy a new digital product in weeks using AI tools, technology ceases to be a differentiator. It becomes a baseline condition of competition rather than a source of advantage within it.

If technology no longer differentiates, the question becomes: what does?

The answer is creativity—not as a cultural value or leadership aspiration—but as an institutional capacity to identify problems that competitors haven’t recognized, to explore solutions before committing to them, and to create products and experiences that reflect genuine insight and not just incremental iteration on what already exists.

This is a demanding capability to build. It’s also a capability that can’t be replicated and commoditized by the AI tools that have leveled the technology playing field.

This isn’t just a banking observation. Andreessen Horowitz general partner Anish Acharya made the same argument in October 2025—in the context of consumer technology, not banking—on the a16z podcast. The episode was titled, aptly enough, “Why Creativity Will Matter More Than Code.”

Acharya’s argument: as AI handles code generation, human value shifts to product sense, storytelling, and what he calls “emotional interfaces.” The ability to define what to build becomes more valuable than the ability to build it.

AI has lowered the barrier to building so dramatically that any individual can now act as a full-stack creator—which means the technical execution gap between competitors has effectively collapsed, and the differentiator is the quality of the idea and the vision behind it.

The question banks—and every company in every industry—must answer then is:
How do we build and create this “creative” ability?

The Four Phases of Creativity

Rick Rubin is the most accomplished record producer in the history of music, with credits spanning Johnny Cash, Jay-Z, Adele, the Beastie Boys, and Metallica. His 2023 book, The Creative Act: A Way of Being, describes how creative work occurs at its highest level and how the conditions that produce it can be deliberately cultivated. The central framework of the book identifies four phases of creative work:

• Seeds are the raw fragments—a chord progression hummed in the shower, a lyric that arrives before the melody, a mood that doesn’t yet have a form.

• Experimentation is where the artist plays with those fragments without commitment or destination—just exploration.

• Crafting is the often-painful process of shaping raw material into something coherent, where choices get made and discarded and made again.

• Completion is release.

Rubin emphasizes that the phases require different mental states and shouldn’t be collapsed. Artists who try to craft before they’ve experimented produce something technically correct but emotionally lacking. Artists who keep experimenting because they’re afraid to commit never finish anything.

The greatest records Rubin produced came from artists who understood which phase they were in and honored it completely. And it’s all applicable to the business world.

The Creativity Framework in Banking

The four-phase framework applies to the banking industry:

1) Seeds. In Rubin’s framework, Seeds are the raw material of creative work—fragments of observation, perception, and experience that have not yet been shaped into anything. In a musical context, these might be an unresolved chord progression, a lyric without a melody, or a mood that has not yet found its form.

In banking, Seeds are the unprocessed signals that exist throughout the institution: the customer complaint pattern that recurs in call center logs without ever reaching a product team, the workaround a small business owner developed because the cash management tool doesn’t accommodate the way her business actually operates, or the life event—a business sale, a divorce, an inheritance—that the product portfolio handles poorly because it was designed around different customer circumstances.

These signals are present everywhere. Branch personnel encounter them in daily interactions, customer service teams document them, commercial relationship managers hear them in client conversations. But the organizational process for systematically collecting it, evaluating it without prejudice, and routing it to decision-makers doesn’t exist.

Rubin argues that the Seeds phase requires gathering observations without filtering them immediately through existing product categories or roadmap priorities. This is difficult for institutions built around regulatory compliance and risk governance.

It’s the capacity, however, that distinguishes institutions with genuine insight from those that bring to market what their competitors launched two years prior.

2) Experimentation. Experimentation, in Rubin’s framework, is the phase in which Seeds are explored without commitment to a predetermined outcome. The purpose is inquiry: whether a problem is accurately understood, whether a proposed solution addresses the right underlying need, whether the customer being designed for is the customer whose behavior will determine whether the product succeeds.

Most bank product and process development processes don’t include an Experimentation phase. They include a requirements phase, which assumes that the problem has already been correctly diagnosed and the solution already defined.

The proliferation of AI has made this gap more important. The speed at which AI-assisted teams can move from concept to prototype has compressed timelines but it also reduces tolerance for the ambiguity that experimentation requires. Banks that use AI to collapse the phases will produce more output, but with less impact.

3) Crafting. Crafting is the iterative work of transforming raw material into a finished product. In banking, this is product development—design, engineering, quality assurance, compliance review, and launch preparation. Banks have invested substantially in the infrastructure to execute this phase: project governance, agile methodologies, development tooling, and vendor management capabilities.

The problem is that financial institutions typical arrive at Crafting before the Seeds have been taken seriously and Experimentation has been completed. Banks that enter Crafting prematurely are executing efficiently toward an insufficiently-examined end.

According to Rubin, this pattern in music produces work that’s technically proficient but lacks the quality that makes it resonate. In banking, the equivalent is a product that performs as specified in the requirements document but generates little adoption.

4) Completion. Completion, in Rubin’s framework, isn’t the achievement of perfection—it’s the release of something that has fully expressed what it set out to express. In banking, this standard has an implication that most institutions resist: it requires honest evaluation of whether a launched product has done what it was intended to do for the customer, not just whether it has met its internal performance metrics or not.

The roadmap pressure to move immediately to the next initiative frequently forecloses this assessment. Products are declared complete when they ship. Whether they work—in the sense that matters to the customer—is a question that often goes unasked.

Marketing: From Interruption to Insight

Go to most community banks’ and credit unions’ websites and count how many times you see the words “trusted,” “local,” “relationship,” and “community.” The words are identical, the imagery interchangeable, the value propositions indistinguishable.

This is what happens when marketing is organized around Crafting without Seeds or Experimentation: the brief goes to the agency, the agency produces executions, the executions get approved by committee, and the result is work that is professionally produced and competitively invisible.

What needs to happen across the phases:

• The Seeds phase in bank marketing should produce genuine observational intelligence about customers—not survey data that confirms what the institution intuitively knows—but insight into how customers actually think about their financial lives and the problems they want solved. The bank that understands that—and builds its marketing around it—operates from a different seed stock than one running another “we’ve served this community for 90 years” campaign.

• Experimentation means testing ideas before committing budgets to them. This isn’t an A/B test of two headlines, but a genuine questioning of whether the audience, the message, the channel, and the moment are correctly identified.

• AI has made the Crafting phase of marketing fast and cheap. Copy, imagery, targeting, and personalization at scale are all more accessible than they were before. What AI can’t do is generate the insight that makes marketing worth producing in the first place. The institutions that use AI to produce more of the same faster will achieve broader distribution of work that nobody remembers.

Product Design: The Lost Skill That AI Won’t Bring Back

Most banks and credit unions don’t do product design—they do product selection, choosing from what vendors and fintech partners have already built, then configuring and co-branding it for their market.

That’s not a criticism. It’s a rational strategy. The economics of custom product design and development doesn’t work at community institution scale, and the vendor and fintech ecosystem has been able to deliver what mass market customers need.

Some institutions have bucked the trend. Vantage West Credit Union, for example, didn’t launch HUSTL because a fintech partner offered a gig-worker banking solution. It was launched because the credit union identified that freelancers, side hustlers, and gig workers cobbled together bank accounts, spreadsheets, and payment apps to manage their financial lives—and nothing in the market addressed their needs.

The credit union’s observation was a Seed. The decision to explore whether they could fill that gap before committing to a product direction was Experimentation. The Build came third.

The question is no longer whether an institution can afford to build something. The question is whether an institution has done enough Seeds and Experimentation work to know what to build in the first place. Most haven’t, because they never developed that capacity—they’ve outsourced the thinking along with the building.

A creative “discipline” is missing in most banks because partner solutions were sufficient. It’s a discipline that AI tools can’t provide, however. AI can build faster once the direction is set, but it cannot determine whether the direction is right. That judgment requires human capacity that most institutions have allowed to atrophy.

The creativity advantage in product development isn’t the capacity to build faster—it’s the capacity to see clearly enough what’s worth building before in the first place.

Customer Service: An Overlooked Source of Creative Intelligence

Customer service is where the consequences of every upstream product, process, and marketing decision eventually surface. It’s also the most systematically underutilized source of creative intelligence in most banking institutions. Across Rubin’s phases:

• Every call center interaction, branch conversation, and chat session is a potential Seed. Customers provide unfiltered, real-time signals where products don’t match needs, where a process creates friction, and where expectations aren’t met. Most banks process this signal as an operational problem rather than as a creative input. The call center exists to resolve contacts efficiently—volume metrics and handle time are optimized, and escalation paths are defined. What’s missing is a mechanism to connect the insights generated in those contacts with the people who can change the products and processes that caused the contacts.

• The Experimentation phase in customer service means testing different approaches to recurring problems before standardizing a response. When the same question keeps surfacing—e.g., about a fee, about an account feature, about how a digital tool works—the first-order response is to train staff to answer it more efficiently. The creative response asks why the question keeps coming up, whether the product design created the confusion, whether the fee structure is generating resentment that won’t show up in attrition data until it’s too late to address, and what a different approach to that customer moment might look like.

• The Crafting phase in customer service is the design of interaction models, escalation protocols, scripting, and training. This is the operational process that determines how contact with customers should go. Banks invest heavily here. The creativity gap isn’t in the execution of service delivery—it’s in the failure to feed what customer service learns back into the Seeds and Experimentation phases of product development and marketing.

• Completion in customer service isn’t contact resolution. It’s the elimination—through product improvement, process redesign, or communication clarity—of the conditions that generated the contact in the first place. A customer service operation becomes more efficient at resolving the same recurring problems has confused Crafting with Completion. The creative standard is higher: did we learn enough from this interaction to make the next one unnecessary?

The institutions that treat customer service as a creative intelligence function—rather than a cost center to be optimized—will find themselves with a continuously refreshed source of Seeds that their competitors are systematically discarding.

AI Agents Won’t Close the Creativity Gap

If AI can accelerate product development, personalize marketing at scale, and handle routine customer service interactions—why can’t it also handle the creative work that precedes all of that? Because:

1) AI doesn’t know what matters. The Seeds phase isn’t a data collection exercise. Banks already have more data than they know what to do with—transaction histories, call center logs, survey responses, digital behavior, application flows. The constraint isn’t the volume of signal it’s the capacity to notice which signals matter, why they matter, and what they might be pointing toward that nobody has yet articulated.

AI agents can scour the Web for relevant news but their ability to know what to do with the inputs—and what the effectiveness of those actions is—could take years and thousands of interactions to develop. Experienced—and creative—humans can figure it out in seconds.

A loan officer who hears a small business owner tell her something that doesn’t fit the standard credit narrative is operating in the Seeds phase when she pauses on that disconnect rather than skipping past it. An AI agent processing the same interaction will classify it, route it, and summarize it according to the patterns it was trained on.

AI agents surface patterns in the data that already exist. The most important Seeds are often the observations that don’t fit the pattern—the anomalies, the exceptions, the customer behaviors that the existing product categories have no language for. Those are the signals that pattern-recognition systems are designed to minimize.

The institutions that develop human capacity to notice what the data can’t will find Seeds that their AI-only-equipped competitors will miss entirely.

2) AI can’t question the question. Experimentation isn’t just testing variations of a defined solution. Its interrogating whether the solution space has been correctly identified in the first place. That distinction is critical—and it is the distinction that puts genuine Experimentation permanently outside the reach of AI agents.

An AI agent given the task of improving small business checking account adoption will generate hypotheses, design tests, analyze results, and recommend optimizations. And it will do it faster and at greater scale than any human team.

What it won’t do, however, is stop and ask whether improving small business checking account adoption is the right problem to work on, whether the bank’s product architecture for small business customers is the constraint, whether the customers being targeted have a need that the product doesn’t address, or whether a completely different framing of the problem might produce a better outcome.

That kind of questioning requires the willingness to question the premise of the assignment. It requires the capacity to tell the organization that it has been asking the wrong question, with all the organizational friction that entails.

AI agents aren’t built to challenge their inputs—they’re designed to optimize within them. That’s useful, but it’s not Experimentation.

The most valuable output of the Experimentation phase is the discovery that the institution is about to build the wrong thing. AI agents won’t produce that—they’ll produce increasingly refined versions of the wrong thing, faster.

3) AI won’t exercise judgment. Crafting is the phase where AI delivers the most value. Design tools, development acceleration, compliance screening, content generation, personalization—the Crafting-phase contributions of AI reshape what bank marketing teams can accomplish with a given level of resourcing.

The risk in the Crafting phase, however, is that it hides the need for human judgment at the moments it needs it most.

Crafting requires a continuous series of decisions about what to keep, what to discard, and what to reconsider altogether. In music, Rubin describes this as knowing when a recording has captured something truly genuine and not just technically proficient.

In banking, it’s the judgment call about whether a:

• product feature genuinely serves the customer vs. serving the institution’s revenue model while appearing to serve the customer;

• marketing message is honest about what the product does vs. optimized to generate response; or

• customer service protocol resolves the customer’s actual problem vs. resolving the institution’s exposure to the customer’s actual problem.

AI agents will make these distinctions according to the objective they’ve been given. If the objective is conversion, the output will optimize for conversion, if the objective is engagement, the output will optimize for engagement.

The judgment about whether conversion or engagement is the right objective—whether the metric being optimized reflects genuine customer value or a proxy for it—is a human responsibility that can’t be delegated to AI.

4) AI can’t evaluate whether any of it mattered. Completion, in Rubin’s framework, is the honest assessment of whether the work has fully expressed what it set out to. In banking, it’s the evaluation of whether the product, campaign, or service model genuinely resolved the customer problem that initiated the effort—not whether it met the metrics that were defined at the project’s outset.

This distinction exposes the limitation of AI in the creative process.

AI agents and tools are great at measurement. They can track adoption, engagement, satisfaction scores, retention rates, and revenue contribution with precision and speed that no human (or team of humans) can match.

What AI can’t do is evaluate whether the measurements matter, whether the correct metrics were correctly chosen, whether the original problem statement was accurate, or whether the customer is really better served or just less likely to complain.

The institutions that close the loop between original customer insight and honest outcome evaluation will do so because of human judgment. AI agents can inform that judgment, but they can’t substitute for it.

Creativity as a Strategic Imperative

In an environment where AI provides every institution access to equivalent technical capabilities at equivalent cost and speed, competitive differentiation will accrue to institutions that are structurally better at: 1) identifying what customers actually need; 2) exploring solution spaces before committing to them; and 3) evaluating their own work against the standard of customer value rather than internal process completion.

That’s the creativity advantage. Unlike a technology advantage, it can’t be acquired from a platform vendor, replicated from a conference presentation, or closed by a competitor with a larger implementation budget.

Rubin spent decades observing the difference between musicians who possessed technical proficiency and musicians who had something worth saying. The distinction determined which work endured. The same distinction will determine which banking institutions build lasting relevance in the decade ahead.

The Creativity Gap Is a Human Opportunity

I’m not arguing that AI has no role in the creative process of banking. My argument is that in the phases where creative work is most important—noticing what matters, questioning the premise, exercising judgment, and evaluating honestly—humans are irreplaceable and where the competitive advantage will be won or lost.

AI agents will get faster, more capable, and more deeply embedded in every operational layer of the institution. That makes the humans who do what AI can’t do more valuable, not less.

The banks that understand this—and invest accordingly in developing human creative capacity alongside their AI infrastructure—will find that the two reinforce each other in ways that neither can achieve alone. Those that treat AI as a substitute for human creativity will discover the limitations the hard way, wondering what went wrong.

At this year’s Financial Brand Forum, Jim Marous and I did an episode of Pardon the Finterruption. During it I said, somewhat nonchalantly:

“The concept of core deposits no longer holds. There are no core deposits. They’re all movable.”

That assertion needs a bit more explanation. Here goes:

Bankers treat core deposits—checking accounts, savings accounts, money market accounts held by “real” customers—as the bedrock of funding stability. Low-cost, sticky, and loyal.

The assumption is simple: Get someone to open a checking account with you and you’ve locked in a funding source that will sit there quietly, earning you a fat net interest margin.

About 10 years ago, I started calling checking accounts “paycheck motels:” temporary places for people to park their money before it moved on the bigger and better places.

Bankers ignored this because they assumed that fintechs—the not-so-bigger places where people were moving their money—were attracting young consumers who didn’t have much in the way of deposits.

A 2025 Cornerstone report titled Stemming the Deposit Outflow: The $2 Trillion Investing Opportunity for Banks and Credit Unions dispelled that notion, finding that 60% of the deposits moved out of banks and credit unions between 2020 and 2025 came from Gen Xers and Baby Boomers.

During the 2022-2023 rate cycle, deposit outflows hit community banks and credit unions hard. Money market funds crossed $6 trillion. Fintechs offering 4-5% APY on savings accounts added millions of accounts in months. Apple Card’s savings account gathered nearly $1 billion in deposits in its first four days.

Core Deposits Are a Myth

The behavioral assumption underneath “core” was always more wish than fact. Three things converged to dispel the wishes:

• Visibility. Consumers can now see exactly what they’re earning—and what they could be earning—in seconds. The information asymmetry that made deposit inertia possible is gone. Every fintech with a high-yield savings account is running ads against it.

• Friction evaporated. Moving money used to take effort. ACH transfers took days. Zelle, instant payments, and same-day ACH mean a consumer can shift $50,000 between institutions in minutes. The switching cost that “stickiness” was built on is approaching zero.

• Relationships fragmented. The average consumer now holds accounts at 2.5+ institutions. The “primary” relationship—where payroll lands, bills autopay, and the checking account lives—is increasingly decoupled from where savings and investments sit. Consumers learned to unbundle. Bankers are fighting it.

Regulations Haven’t Caught Up

Believe it or not, regulators move slower than bankers do. The behavioral shift in deposits is running up against how regulators classify, price, and manage deposit risk.

The entire US bank regulatory framework for liquidity was built on the same inertia assumption bankers made. Core deposits—operationally defined as non-maturity deposits from retail customers—are treated as stable funding. The premise: retail depositors don’t run. Institutional money runs, but retail money stays.

Silicon Valley Bank blew that premise out of the water in 48 hours.

What happened at SVB wasn’t just a concentrated depositor base problem. It was a demonstration that digital banking infrastructure, social media, and a highly networked depositor community can produce a bank run at a velocity the regulatory framework never imagined.

The FDIC, the Fed, and the OCC were all operating on assumptions baked into rules written before mobile banking existed at meaningful scale. SVB lost $40 billion in deposits in a single day. Traditional bank run models didn’t take that number into account.

The regulatory response since then has been revealing in what it shows about the gap.

The FDIC assumed that uninsured depositors were more sophisticated and more mobile. But the SVB postmortem also showed that even insured depositors behaved differently than the models predicted when panic set in.

Brokered deposit rules are another problem: The FDIC tried to modernize a framework that hadn’t been updated since the 1980s when “brokering” meant a clearly identifiable third party physically moving deposits between institutions.

Today, fintech deposit marketplace platforms like Raisin, StoneCastle, ModernFI, Ampersand, and IntraFi blur the intermediary question. When a consumer opens an account through a fintech app and their deposits are swept to a network of partner banks via a middleware layer, is that brokered?

The regulatory answer has been inconsistent, and the ambiguity creates risks for community institutions competing against deposit products that aren’t being assessed and regulated the same way theirs are.

And yet another contributor to the regulatory mess: The Fed suspended the six-withdrawal-per-month limit on savings accounts in April 2020—a rule that had been in place since 1933—and never reinstated it.

The rationale was pandemic convenience, but the practical effect was removing a regulatory friction that had reinforced deposit inertia for nearly a century. Savings accounts are now functionally equivalent to checking accounts in terms of access.

So much for the theory of “deposit stability.”

The CLARITY Act Compromise That Isn’t

The most revealing thing about the CLARITY Act negotiations isn’t what ended up in the bill. It’s what the fight tells us about what banks actually believe about deposits.

Here’s a quick recap of where things stand: The GENIUS Act, signed into law in 2025, established a regulatory framework for stablecoins and prohibited stablecoin issuers from paying yield directly. Banks celebrated.

Then they noticed the loophole: nothing in the GENIUS Act stopped third-party platforms like Coinbase from paying rewards on user stablecoin balances. The battle shifted to the CLARITY Act, the broader digital asset market structure bill that passed the House in July 2025 and has been stuck in Senate negotiations ever since.

The compromise text released last week by Senators Tillis and Alsobrooks prohibits crypto firms from paying any form of interest or yield on stablecoin balances in a manner that is “economically or functionally equivalent to the payment of interest or yield on an interest-bearing bank deposit.”

Crypto firms can still offer rewards, but only tied to actual usage—transactions, payments, trading activity. The agreement forces a structural shift from a “buy and hold” rewards model to a “buy and use” model.

The banking lobby (ABA, BPI, ICBA) publicly rejected the drafted language anyway, arguing it still threatens the foundation of low-cost deposit funding (they’re right), warning that yield-earning stablecoins could reduce loans by 20% or more.

The Math Ain’t Mathing Here

Banks are arguing that a competing product which pays yield on digital dollars would drain their deposit base by 20%. They’re making this argument, however while simultaneously claiming that their core deposits are stable, loyal, and rate-insensitive.

They can’t have it both ways. Either deposits are sticky—in which case, why the panic?—or they’re rate-sensitive and mobility-prone, in which case the core deposit framework is exactly the fiction I’m arguing for here.

The regulatory outcome from the CLARITY Act won’t resolve anything.

The bill would give regulators a year to publish rules clarifying when companies can reward stablecoin users. That means another year of ambiguity, and another year of community banks and credit unions watching the landscape shift while waiting for regulatory clarity that always arrives late.

I asked Claude to check whether the prediction markets were pricing the odds of the CLARITY Act becoming law in 2026. Current odds: more than 60% (per Claude).

USDC in circulation reached $75.3 billion at the end of 2025, up 72% from 2024. That money had to come from somewhere. Some of it came from institutional and international flows. But some of it had to come from people who decided that holding dollars in a stablecoin—even one that doesn’t officially pay yield—was better than holding those dollars in a savings account earning 0.01%.

Stablecoins are becoming a legitimate part of the consumer financial stack. The faster that happens, the faster the theory of “deposit stability” erodes.

What Community Banks and Credit Unions Should Do Now

For community financial institutions, the regulatory implications cut two ways.

If the regulators update liquidity frameworks to reflect that retail deposits are more mobile than assumed, capital and liquidity requirements will shift, increasing the cost of regulatory compliance.

On the other hand, institutions that update their internal models now—stress-testing deposit outflow scenarios that regulators haven’t mandated yet—could get ahead of the curve.

Community banks and credit unions need to:

• Stop managing deposits as if inertia is a strategy. Inertia is a temporary condition, not a competitive advantage.

• Deepen product relationships. The institutions that performed best in the last rate cycle weren’t the ones with the most “core” deposits. They were the ones with the deepest product relationships—loans, treasury management, business services, merchant services—products that create real switching costs because the operational integration is real.

• Do better deposit segmentation. Not all deposits behave the same way. Rate-sensitive money needs to be identified and priced differently than truly relationship-anchored deposits.

You can’t wait for the regulatory framework to catch up to reality. That’s never happened before—why would you expect it to happen now?

The concept of “core” deposits was useful for regulatory capital analysis, but less so (or not at all) as a strategic framework. The institutions that replace the core deposit “concept” with actual data about customer behavior, relationship depth, and rate sensitivity will out-execute those who talk about core deposits like it’s 2015.

If you want to understand the current state of AI in banking, look at the FIS-Anthropic announcement like it’s a Rorschach test. What you see reveals more about you than about the deal.

FIS announced a partnership with Anthropic to build AI agents for banking, starting with a Financial Crimes AI Agent that compresses AML investigations from hours to minutes.

The agent automatically assembles evidence across a bank’s core systems, evaluates activity against known typologies, and surfaces the highest-risk cases for investigator review.

Picking financial crimes as the first use case isn’t surprising. Banks spend billions on anti-money-laundering efforts to satisfy rigorous federal mandates and regulatory oversight. AML is the perfect starting point for an AI agent rollout:

• the ROI is concrete and measurable;

• the regulatory pressure is relentless; and

• no bank is going to brag that it skimped on catching drug traffickers and terrorists.

But let’s not confuse a smart wedge for a transformation.

The “Humans Still Decide” Hedge

FIS noted that the decisions about cases will still be made by human investigators. According to Anthropic: “every decision stays with the investigator.”

This caveat will appear in every single AI-in-banking press release for the next three years, so we might as well start tracking how long it actually holds.

The honest translation is: we’re automating the evidence assembly and triage, not the judgment call. Which is still valuable. Investigators don’t need better judgment—they need fewer cases that don’t deserve their attention, and that’s what the agent delivers.

In theory, at least.

The Embedded Engineers Model

Another detail worth paying attention to is how Anthropic will deploy its people. Anthropic’s Applied AI team and forward-deployed engineers (or FDEs as I learned today) will be embedded with FIS to co-design the Financial Crimes AI Agent and transfer knowledge so FIS can build and scale additional agents over time.

That’s the Palantir playbook: Embed your best engineers, build the first thing together, then hand FIS the keys to build the next ten agents on their own.

AML is just the opening act. Beyond financial crimes, FIS intends to develop more agents with Anthropic for credit decisioning, deposit retention, customer onboarding, and fraud prevention.

The product ambition is to become the agentic infrastructure layer for banking operations. That’s a genuinely interesting strategic position—if they can execute it.

A New Strategy For Ginning Up Your Stock Price

FIS stock, which has seen a roughly 40% decline over the past twelve months on AI-disruption fears, popped 7% overnight. A press release with two pilot banks (BMO and Amalgamated) moved the stock 7%.

What Community Banks Should Actually Think About

Here’s the question that didn’t appear in a single press release or analyst note:

What does this mean for the 3,000+ community banks and credit unions that run on FIS platforms?

FIS CEO Stephanie Ferris said “every bank in the world wants AI that acts, not just assists.” She’s right.

But “every bank in the world” and “BMO and Amalgamated Bank” don’t belong in the same sentence. BMO is a top-25 North American bank with the IT staff and compliance infrastructure to absorb a new agentic workflow. Amalgamated Bank is a progressive-leaning institution with $9 billion in assets and a tech-forward posture. Neither is a $500 million community bank in rural Ohio.

The relevant questions for community bank executives aren’t “is this technology real?”they’re:

• When does it reach the standard FIS product tier, and at what price point?

• What core system integrations are required to assemble the evidence the agent needs—and do community banks have the data infrastructure to support it?

• Who owns the liability when the agent surfaces a false positive that triggers a regulatory action?

• Why Anthropic and not some other AI provider that isn’t on the government’s list of national security threats?

The Fintech Snark Tank Take

The market reaction to this deal tells you something important that the deal itself doesn’t.

The emergence of advanced AI models has recently pressured the valuations of legacy software providers like FIS, whose stock has declined over 25% this year. Investors are concerned that traditional software products could become obsolete as firms build their own internal AI solutions.

However, the FIS/Anthropic partnership suggests that AI labs may instead accelerate existing enterprise software through strategic industry-specific integrations.

That’s the plot twist here. The market had been pricing FIS as a casualty of AI disruption. This deal reprices it—at least temporarily—as a potential beneficiary. The logic: Anthropic needs distribution and regulated data environments; FIS has both. Rather than compete, they embed.

Whether that thesis holds depends entirely on execution. AML investigations moving from hours to minutes is a real outcome if it materializes. Two pilot banks in H2 2026 isn’t a transformation—it’s a proof of concept.

Wall Street gave FIS a 7% trophy for announcing a proof of concept. The banking industry should wait to see what it scores on the actual test.

Thanks to the passing of the GENIUS Act last year, stablecoins are a board- or executive team-level topic of discussion in 63% of banks. Nearly 1 in 10 will invest in or deploy stablecoin-related capabilities in 2026.

Half of the banks surveyed by Cornerstone Advisors said deposit retention or generation is driving their interest in stablecoins, with competitive threats from fintechs and crypto firms also cited as a strong influence.

Despite the nascent nature of stablecoins in the US, 4 in 10 bankers believe that tokenized money—e.g., tokenized deposits, stablecoins—will become a common utility, similar to ACH or debit rails.

The US Department of the Treasury estimates that stablecoins could displace “up to $6.6 trillion” in deposits currently held in US banks if they offer interest.

“Up to $6.6 trillion” isn’t a helpful estimate for community banks, however. They need to know where, when, and through what channels stablecoins will siphon value from banks—and what to do about it—now.

The Three Layers of Stablecoin

Stablecoins aren’t speculative cryptocurrencies. They’re dollar claims packaged in a programmable, interoperable, and globally accessible wrapper. And thanks, in part, to the GENIUS Act, many banks think about stablecoins from just a balance-sheet perspective. But stablecoin stability and risk plays out across three layers:

1. The Balance Sheet Layer

The GENIUS Act addresses the question: Does a stablecoin issuer hold enough high-quality assets to be solvent if redemptions are demanded? The answer: Mostly yes, but a capital buffer analysis reveals some issues and ironies.

The issuers most enabled by the GENIUS Act—e.g., Circle, PayPal, Paxos—earn Treasury rates on their reserve assets, pay nothing to stablecoin holders (GENIUS prohibits interest payments on stablecoins), and pocket the spread. That spread funds operations and builds capital.

The problem is that T-bill rates minus operating costs produces razor-thin equity cushions. Circle’s leverage ratio hovers around 0.1 to 0.2%. Under bank standards that’s not just undercapitalized—it makes them candidates for receivership.

The irony is that the safer the backing asset (shorter duration, higher quality), the lower the yield, the thinner the spread, and the thinner the capital buffer.

An issuer holding Bitcoin and corporate bonds like Tether generates more excess capital because those riskier assets yield more. The regulation creates a perverse incentive structure where the most GENIUS-enabled issuers are the most capital-fragile.

By itself, balance sheet fragility doesn’t necessarily kill a stablecoin. A 0.2% leverage ratio is terrifying for a bank because bank assets are illiquid loans that can’t be rapidly monetized. In normal conditions, a stablecoin issuer holding overnight repos could liquidate its entire balance sheet in 24 hours.

The balance sheet layer, however, interacts with the plumbing layer, and that’s where the real problem lives.

2. The Plumbing Layer

The plumbing question is: Even if the balance sheet is solvent, can the redemption mechanism actually function under stress? The answer: Not reliably, and for a structural reason that isn’t going away.

A redemption requires the stablecoin issuer to sell Treasuries to get bank deposits to pay the redeemer. That sale goes through a broker-dealer who has to take the Treasuries onto its balance sheet (at least temporarily), which pushes its supplementary leverage ratio toward its regulatory floor.

When dealer banks are already operating near that floor, their capacity to absorb incremental Treasury selling is severely constrained.

The worst Treasury market disruption in decades involved a net selloff of less than $100 billion. Current stablecoin market cap is roughly $230 billion. Even a modest redemption run of 15% to 20% of outstanding stablecoins approaches that amount. And a run wouldn’t likely happen in isolation—it would happen during a period of broader market stress when dealer balance sheets are already under pressure.

Here’s why the plumbing layer is so important: People (i.e., consumers, businesses, institutions) will hold stablecoins because they don’t have to worry about the underlying mechanics. The moment those mechanics become visible or unstable, confidence will collapse.

3. The Technology Layer

The technology layer asks: Even if the balance sheet is solvent and the plumbing can process the redemption, can the actual blockchain transaction execute reliably? The answer: Sometimes no, for reasons entirely outside the issuer’s control.

Smart contract risks—logic bugs, upgrade vulnerabilities, bridge failures, key mismanagement—are real.

When Paxos accidentally minted $300 trillion of PYUSD in October 2025 its financial reserves were fine, the plumbing was intact, but a software error briefly destroyed confidence and broke the peg anyway. That’s a technology layer failure.

A recent report from MIT makes a case that as stablecoin volume grows, the economic incentive to attack the underlying blockchain grows faster than the cost of mounting the attack.

The reward for a successful attack scales with transaction volume—but the cost doesn’t. This creates a widening gap between attack incentive and attack deterrence that’s intrinsic to the architecture, and not solvable through better issuer practices.

What Community Banks Need to Know

If you only evaluate stablecoins through the balance-sheet lens, you miss the risks that show up when markets are stressed or systems fail. A stablecoin can be “fully backed” and still fail if redemption plumbing clogs or if the technology layer breaks.

What bankers need to know:

1. Stablecoins don’t just “steal deposits.” They transmit liquidity stress. Even if a bank never issues a stablecoin, token-driven flows can create sharper day-to-day liquidity swings. If a bank partner (or customer) operates in a stablecoin ecosystem—directly or through a fintech—the bank could inherit a set of balance-sheet dynamics that look a lot like “hot money,” because redemption behavior can change quickly and at scale.

This isn’t theoretical. Stablecoins are built for speed. The very feature that makes them attractive—instant movement—also means outflows can accelerate faster than traditional deposit runoff patterns.

2. Stablecoin risk is as much about market plumbing as it is about reserves. Redemption at par depends on more than holding safe assets. It depends on the ability to convert those assets into cash under stress, which can mean reliance on Treasury market liquidity, repo markets, dealer balance sheets, and operational capacity at intermediaries. In good conditions, this all looks smooth. In stress, it can get noisy fast.

Community bank takeaway: you can’t evaluate stablecoin exposure the way you’d evaluate a money market fund. Stablecoins introduce a blend of liquidity, market microstructure, and operational risks that most banks aren’t staffed to underwrite.

3. The biggest threat is interception not replacement. Stablecoins don’t have to replace deposits to hurt community banks. They just need to capture payments and settlement to push banks into a commodity role as on/off ramps and compliance wrappers while value accrues to wallets, platforms, and networks.

The parts of the bank value chain where stablecoins can intercept economics include merchant settlement and e-commerce payouts, cross-border transfers and remittances, B2B supplier payments and platform payouts, disbursements (e.g., gig workers, insurance claims, payroll), and treasury “operating cash” that becomes more mobile and programmable.

If stablecoins become a default rail for moving dollars, the risk for community banks is being relegated to slow, expensive plumbing that customers use only because they must—not because they want to.

What Community Banks Need to Do (Now)

Community banks can’t assume that “we’re not in crypto” equals “we’re not exposed.” They can be exposed through their customers’ behavior, vendors’ integrations, and their own rail economics.

Every bank—big or small—needs a stablecoin strategy (now). That’s not a choice. The choice is to what extent do you play offense or defense. For both approaches, a bank’s stablecoin strategy should address the three layers described above:

1) Balance sheet actions need to defend—and price for—volatility. This means:

• Treating stablecoin-adjacent deposits as hot money. If you bank fintechs, processors, exchanges, or large digital platforms, assume higher runoff risk and manage it accordingly.

• Updating liquidity stress tests for token-speed outflows. Model intraday spikes, rapid outflows, and correlated payment demands.

• Setting concentration limits and triggers. If stablecoin-linked balances exceed a threshold, require additional buffers and senior-level review.

• Repricing treasury relationships for liquidity intensity. If a client drives volatile settlement and redemption flows, get paid for the liquidity burden.

2) From a plumbing layer perspective, know where dollars go when tokens move:

• Map exposure across rails. Identify which clients are sending/receiving funds to on/off ramps, and where token-linked flows show up in wires, ACH, RTP, and FedNow.

• Demand a “redemption dossier” from partners. If a fintech partner touches stablecoins, require clarity on authorized redeemers, redemption cutoffs, intraday funding needs, liquidity backstops, and stress contingencies.

• Compete on settlement optionality. Banks can’t afford to be slow, opaque, and expensive. Even a defensive strategy requires better real-time payments capability and a treasury experience that doesn’t make clients look elsewhere.

3) From a technology perspective:

• Treat stablecoin rails as regulated third-party technology. If you integrate to token networks—directly or via vendors—your risk program must cover smart contract risk, custody/key management, governance/upgrade controls, monitoring, incident response, etc.

• Be “token-aware.” Few community banks will need to mint their own stablecoin. What they need to do is help customers accept token-based payments, convert/settle efficiently, provide compliant reporting, and offer treasury tools that make stablecoins unnecessary for most domestic use cases. And that means modernizing settlement speed and transparency.

Stablecoins won’t rock community banks overnight. But they will pressure deposit economics at the margin, reroute payments value, and introduce new liquidity and operational risks, especially for banks that—directly or indirectly—touch the stablecoin ecosystem.

You’re forgiven if you don’t know who Rick Rubin is, but telling me that you don’t who he is puts you on my “I’m not sure we can be friends” list.

For the uninitiated, Rubin is the greatest American record producer ever, working with artists like the Beastie Boys, Red Hot Chili Peppers, Johnny Cash, and Adele across hip-hop, rock, metal, and country genres. He’s a nine-time Grammy winner and was named one of TIME’s 100 most influential people.

He’s also the author of the most important business book of this decade. Except that nobody but me thinks his book is a business book.

Google describes The Creative Act: A Way of Being as “a guide to viewing creativity not just as making art, but as a daily, present, and open-minded way of living. It emphasizes nurturing curiosity, cultivating self-awareness, and focusing on the process rather than the results.”

That’s all well and fine, but as far as I’m concerned it’s a guide to reinventing business organizations in the age of AI and finding the next source of competitive advantage and differentiation.

This post will focus on how it applies to the management consulting industry (which so many self-appointed luminaries on LinkedIn think is dying because of AI).

Technology is No Longer a Source of Competitive Advantage

For the better part of three decades, I (and plenty of other people) have considered technology to be a source of competitive advantage.

Having a vision for how technology can create new products/services or change the way things get done—and then investing in, deploying, supporting, and scaling that vision has helped many firms create differentiated competitive advantages.

That’s over.

When anyone can spin up a working application in an afternoon using AI, when any firm can stand up a data pipeline or automate a workflow in hours rather than months, when the technical execution gap between a boutique shop and a global firm has effectively collapsed—technology stops being a differentiator.

Which raises an uncomfortable question for industry after industry that has organized itself largely around technical capability: If technology doesn’t create a competitive advantage or differentiate you, what does?

Rubin’s answer—though he’s writing about music, not management consulting—is creativity. Not as an aesthetic sensibility, but as a discipline—a practiced capacity to see problems differently, to ask questions nobody else is asking, and to produce recommendations that couldn’t have been generated by prompting an AI model.

The Phases Problem Nobody in Consulting Talks About

Rubin structures creative work around four phases: Seeds, Experimentation, Crafting, and Completion. In a music context:

• Seeds are the raw fragments—a chord progression hummed in the shower, a lyric that arrives before the melody, a mood that doesn’t yet have a form.

• Experimentation is where the artist plays with those fragments without commitment or destination—just exploration.

• Crafting is the often painful process of shaping raw material into something coherent, where choices get made and discarded and made again.

• Completion is release.

Rubin emphasizes that the phases require different mental states and shouldn’t be collapsed.

Artists who try to craft before they’ve experimented produce something technically correct but emotionally lacking. Artists who keep experimenting because they’re afraid to commit never finish anything.

The greatest records Rubin produced came from artists who understood which phase they were in and honored it completely.

This process applies to a typical consulting engagement.

A partner sells a project with a hypothesis (seed) already baked in. Associates spend week one building the work plan and week two populating the slide shell. The Crafting phase has swallowed the Experimentation phase before anyone has spent serious time with the client’s actual problem.

The result is a recommendation that confirms what the consulting team believed on day three, packaged in a 90-slide deck that looks authoritative, and is packed with conclusions that don’t surprise anyone who’s been around for more than a few years.

Here’s what makes this relevant now: AI can execute the Crafting phase faster than any team of consultants. It can synthesize, structure, format, and populate.

What it can’t do is execute the Experimentation phase because that requires genuine curiosity about a client’s specific situation—not pattern-matched outputs from training data.

Consulting firms that learn to protect the Experimentation phase will have something to that differentiates them because AI can’t replicate it. The consulting firms that use AI to collapse the phases even faster than they already were will simply produce bad work at higher speed.

The Four Phases Applied: What it Looks Like in Practice

Abstract frameworks are easy to endorse and hard to act on. So let’s make this concrete.

Cornerstone works with banks and credit unions on important technology decisions—core system replacements, digital banking platform selections, loan origination system overhauls, and contract negotiations that shape the institution’s cost structure and operational flexibility for the next decade.

These are high-stakes, high-visibility projects. They’re also projects where the four-phase discipline Rubin describes is routinely violated in ways that are entirely predictable and entirely avoidable.

1) Seeds in a technology selection project are the honest, unfiltered observations about why the current situation isn’t working—not the sanitized version that gets written into the RFP.

The core system that the ops team has built seventeen workarounds for because it can’t handle the institution’s actual product set. The digital platform whose mobile app ratings have been declining for two years. The loan origination system that the commercial lenders refuse to use because it was designed for consumer lending and nobody wants to admit the implementation was a mistake.

Most institutions arrive at a vendor selection project with a problem statement that has already been negotiated into palatability. The Seeds phase requires going back upstream—talking to the people closest to the friction, listening without a predetermined solution in mind, and collecting an unvarnished picture of what the institution actually needs before anyone opens a demo environment.

2) Experimentation in this context means interrogating the solution space before an RFP goes out. It means:

• Asking whether a full core replacement is actually the right answer, or whether a modern middleware layer solves the real problem at a fraction of the cost and disruption.

• Exploring whether the digital banking gap is a platform problem, an integration problem, or a product design problem.

• Stress-testing the assumption that a new loan origination system will change behavior if the underlying credit culture and process haven’t changed first.

These are uncomfortable questions to raise once a project is already scoped and staffed.

The Experimentation phase is where a consulting firm earns its fee—not by running a more efficient process, but by helping the client ask better questions before the process starts.

3) Crafting is the selection and negotiation process itself. This is where most of the visible work happens, and where most institutions focus their attention and measure the engagement’s value. Cornerstone’s work in the Crafting phase follows a well-developed and thought-out methodology.

The risk isn’t in the execution—it’s getting to this stage with a problem statement that isn’t honest and a set of requirements that reflect what the institution has always done rather than what it’s trying to become.

4) Completion in a technology selection project isn’t a signed contract. It’s the point at which the client is confident that the selected vendor and solution, implemented as planned, will resolve the real problem that initiated the project and not just the problem as it was written into the RFP.

The consultants and clients that rush to Crafting—that treat the RFP as the starting line rather than a midpoint—can find themselves three years into an implementation wondering how the selection process produced this outcome.

Rubin would recognize the pattern immediately—it’s the album that sounds great but leaves the listener unmoved. In consulting, it’s the engagement that checked every process box but missed the point entirely.

The Neutral Witness: The Most Underused Skill

Buried in the middle of the book is a concept Rubin calls the “Neutral Witness”—the ability to observe your own work without ego investment. Asking what the work actually needs rather than defending what you wanted it to be.

Most consultants oscillate between two failure modes. Either they’re too attached to the hypothesis that got the engagement sold, or they’re so focused on managing the partner’s or client sponsor’s expectations that every finding gets softened before it sees the client.

The Neutral Witness is the third position—internal editorial distance that makes honest assessment possible.

Rubin ties this to meditation practice, which doesn’t map cleanly onto a consulting engagement workflow. But the underlying mechanism does.

When you train yourself to observe your analysis without immediately defending it, you start asking better questions. You notice when the cost model doesn’t support the operating model recommendation, or when the benchmarks compare things that shouldn’t be compared. You ask what the engagement needs rather than what the deck needs.

AI has no Neutral Witness problem—or capacity. It doesn’t get attached to its outputs. It also doesn’t care whether it’s true or right.

The judgment about whether a recommendation is genuinely sound or just plausible requires a human with enough intellectual honesty to apply real scrutiny. That’s a creative act. It’s also the consulting act that matters most.

Self-Doubt Is Data

Here’s the Rubin argument that translates most directly to consulting work: self-doubt isn’t an obstacle. It’s a signal.

When something feels wrong about an analysis—the segmentation, the attribution logic, the way a causal claim is being made from correlational data—that nagging sense is often the most accurate quality signal you have.

Most consultants either suppress it under deadline pressure or surrender to it entirely and conclude the whole workstream needs to be rebuilt from scratch at 11pm the night before the client readout.

Rubin argues for a third response: listen to it. Figure out what it’s pointing at.

The discipline he’s describing—taking the doubt seriously without being paralyzed by it—is rarer in professional services than it should be. Consulting culture tends to reward confidence.

The person who sounds certain in the room gets airtime. The person who says “I’m not sure this logic holds” gets managed out of client-facing work.

[Note: the paragraph above completely describes why the author of this post has failed miserably as a consultant over the past 40 years]

Here’s the AI-era version of this problem: when the doubt arrives about AI-generated analysis, there’s a new temptation to suppress it on the grounds that the model is probably right and the instinct is probably wrong.

Deferring to the AI output over the human judgment is backwards. The model doesn’t know what it doesn’t know. The experienced practitioner does. That expertise, expressed as productive doubt, is what clients are paying for. Surrendering it to the tool is surrendering the differentiator.

The Quality of the Lens: The New Competitive Advantage

Rubin never mentions AI. But his core argument is a direct response to the moment the management consulting industry is actually in right now.

When any consultant with a Claude subscription can generate a competitive landscape, a market sizing, or a process improvement framework in forty minutes, what’s the actual value of a consulting engagement?

When AI can synthesize interview notes into a findings summary before the project manager has finished their debrief, what are firms actually selling?

Today—in 2026—these are productivity hacks for consultants and consulting firms. By 2030, they’re standard practice.

Rubin’s answer is the “quality of the lens”: The accumulated judgment about which questions actually matter, which findings are genuinely surprising versus superficially surprising, which recommendations are too convenient, and what the client is not asking that they should be.

No AI tool has that. Tools have pattern recognition across large datasets. Good consultants have points of view built from experience.

This is the reframe the consulting industry needs.

The old consulting competitive advantage: helping clients create a vision for how to use technology to create competitive advantage.

The new consulting competitive advantage: helping clients create a capability to be creative in the application of technologies that everyone has equal access to.

The consulting firms—and consultants—that understand this will restructure their value proposition. The firms that respond to AI by doubling down on technical capability—more tools, more platforms, more certified practitioners—are running toward a cliff.

The Beginner’s Mind Problem at the Principal Level

One of Rubin’s direct borrowings from Zen philosophy is the concept of Beginner’s Mind—the idea that experts see few possibilities while beginners see many.

His argument isn’t that expertise is bad. It’s that the most dangerous thing a professional can develop is competence without curiosity.

This maps well onto how consulting careers progress. The longer you’ve been in the industry, the faster you pattern-match new client problems onto existing frameworks.

Operational inefficiency always traces back to five root causes. Digital transformation always stalls at three organizational barriers. Post-merger integration always surfaces four cultural dynamics. Or whatever number is the right number.

The frameworks aren’t wrong—they’re built from real engagements. But they become constraints when people stop questioning whether they actually apply.

In an AI environment, this problem multiplies. An LLM is also pattern-matching—and doing it faster. The senior consultant’s pattern-matching adds nothing.

The value comes from someone who stops the pattern-match and asks a question the framework doesn’t have an answer for. That’s a creative act. It requires Beginner’s Mind. And it requires actively protecting that capacity against the pressures of expertise and AI-generated plausibility.

Creativity: What it Means for Consulting

Rubin wasn’t a musician, but he spent decades in studios watching the difference between artists who had technical skill and artists who had something to say. The ones with something to say made the records that lasted.

I’m not a good consultant, but I’ve spent decades in the consulting business observing the ones who are really good. Here’s what I know:

• The good consultants aren’t going to be replaced by AI, and using AI to be more productive isn’t going to make a mediocre consultant a good one.

• The good consultants will be those that understand that the 4 phases require different mental states. The great consultants will be those that are able to put themselves into those mental states.

• The good consultants will be those that help their consulting firm be more creative. The great ones will those that help their clients become more creative.

The cross-functional relationship data wasn’t any brighter: only half of CMOs said Marketing’s relationship with the CIO and IT was strong, and just 42% said the same about the CFO and Finance.

Marketing can’t build better attribution without IT. Marketing can’t make the case for budget increases without Finance. A CMO who doesn’t have strong relationships with the CIO and CFO can’t be a strategic leader.

Operationally, Marketing gets called in after decisions about the product, the rate, and the target market have already been made. When Marketing’s role is defined that way, impact gets measured by activity, not outcomes.

To address this, bank and credit union CMOs need to get 3 things in order:

1️⃣ 𝗧𝗵𝗲 𝗴𝗿𝗼𝘄𝘁𝗵 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆. Most Marketing departments allocate budgets to discrete objectives—acquisition, retention, product campaigns—as if each one were independent. They’re not. They’re all expressions of a single question: what are we selling and who are we selling it to? Until Marketing builds and owns the framework that connects those objectives into a coherent growth strategy, it will keep getting pulled in seventeen directions by whoever had the last meeting with the CEO.

2️⃣ 𝗧𝗵𝗲 𝗱𝗮𝘁𝗮 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆. In a Cornerstone study that benchmarked data quality across functional areas—credit, operations, strategy, and marketing—Marketing scored lowest. Closing that gap means treating marketing infrastructure as part of the enterprise data strategy—not a collection of standalone tools that don’t talk to the core.

3️⃣ 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴’𝘀 𝗰𝗼𝗿𝗲 𝗰𝗼𝗺𝗽𝗲𝘁𝗲𝗻𝗰𝘆(𝗶𝗲𝘀). Without defined competencies, Marketing has no basis for pushing back on bad ideas. Every new channel gets tested because nobody can explain why not. A department that has identified measurement and analytics as its core competency—or brand, or creative, or partner ecosystems—has a decision filter. Every channel, every campaign, every vendor relationship gets evaluated against a single question: does this let us do our best work?

The marketing budget and channel effectiveness data in Cornerstone Advisors’ new report is worth a read—but the findings only matter if Marketing has the organizational standing to act on them.

https://resources.fintelconnect.com/financial-marketing-roi?utm_campaign=rs

Digital bank Revolut reported record annual profits and growth in 2025, with year-over-year revenue growth of 46% and an increase in pretax profit of about 55%. The company said its growth came from multiple income streams, particularly card payments, subscriptions, and foreign exchange.

This comes on the heels of Revolut’s application for a US national bank charter with the OCC and FDIC.

Before the fintecherati proclaims (yet again) the decline of the existing banking market, let’s slow down and ask two questions nobody in the fintechosphere seems particularly interested in: 1) What could actually derail Revolut in the US? and 2) What does Revolut need to do—not announce, not plan, but do—to make it here?

What Could Derail Revolut in the US

It’s worth noting that the US has humbled a few well-capitalized fintechs that achieved scale in other parts of the world. Monzo ended up shelving its US ambitions, N26 ultimately exited the market, and Starling never seriously tried. What’s Revolut up against as it plans to enter the US?

1) Low consumer awareness. Revolut’s own US leadership admitted it publicly: “No one knows us in the US.” That’s not entirely correct, though. In a 2025 study I conducted, I found that about 0.5% of Americans—somewhere between 900k and 1 million—had a Revolut account. But, for a company that’s been operating here for years and has 70 million customers globally, that number is embarrassingly low.

2) A crowded battlefield. Revolut is entering a US market where Chime has 22 million customers, Cash App has embedded itself into the financial lives of younger and lower-income Americans, and SoFi already has a charter and a membership model with real traction.

3) A (potentially) tough OCC charter process. Historically, the OCC process is slow, rigorous, and unforgiving on compliance. Revolut has had regulatory headaches in other markets, and didn’t succeed in its attempt to get a state charter in California. The company can’t assume that a more-permissive regulatory environment under the current administration will translate into a fast-tracked charter for a foreign-headquartered fintech with a complicated compliance past.

4) We’re not a “super-app” country. Revolut’s strategy in other countries is being a financial super-app—trading, crypto, budgeting, travel eSIMs, spending analytics, subscriptions, all in one place. Americans have a deeply entrenched habit of assembling their own fragmented financial stack. Replacing it with a single app—from a brand Americans haven’t heard of—is an uphill battle.

5) The US isn’t Europe. Revolut’s playbook worked in European markets where it could leverage digital-native consumers, cross-border pain points, and a more or less homogeneous regulatory environment. The US is different: 50 states with different regulatory nuances, and a consumer base that has more options and lower brand switching incentives.

What Revolut Needs to Do to Succeed in the US

What does Revolut offer the American consumer that they can’t already get? FX-free travel spending? There’s Wise. Crypto trading? There’s Coinbase, Robinhood, and a dozen others. A premium metal card? Every fintech has a metal card now.

Don’t write Revolut off completely just yet, though. There is a path and strategy for how it can succeed here:

• Own a segment. Revolut needs to lead with a specific use case where it’s truly differentiated. International money movement is an obvious candidate, as it’s the primary use case for the Americans who already use the fintech. But there are other under-served segments, for example, gig workers (and creators and side hustlers). Others have tried to serve this segment but have fallen short. Can Revolut succeed here?

• Leverage the charter. A US bank charter means FDIC insurance, direct access to Fedwire and ACH, and an ability to offer credit products without having to rely on a partner bank. Those are meaningful advantages—but only if Revolut has a lending strategy that makes sense for the US market (more specifically, the market segment it owns). Revolut needs a segment-specific hook or pricing strategy to reach the niches that the big guys don’t address.

• Excel at customer service. Revolut’s reputation for not-so-great customer service is no secret. Europeans tolerated it because the product benefits were significant enough. Americans won’t be as forgiving. Before scaling customer acquisition, Revolut needs to be confident that its service infrastructure can handle American volume and American expectations.

• Start marketing now. The OCC charter process usually takes two or more years. Two more years of low brand awareness in the US market, while competitors consolidate their positions, isn’t a luxury Revolut can afford. It needs to build brand equity now—not with a global campaign about “money possibilities” that might play in other countries, but with targeted, differentiated messaging that answers the question every prospective US customer will ask: “Why Revolut?”

If Revolut Can Make it There, Can it Make it Here?

Revolut has advantages that failed the European fintechs who came before it didn’t have: more capital, a more complete product, and a regulatory environment that may be incrementally more accommodating.

Whether those advantages are enough depends almost entirely on whether they’re willing to treat the US as the distinct, difficult, and relationship-driven market it actually is, rather than just a checkbox on the path to 100 million customers.

For four years Elon Musk has been telling us that X Money would become a super app and a payments powerhouse. He first floated the idea in a 2022 investor pitch deck and promised a launch by end of 2024 (and then 2025). Finally, X Money is making its debut in early public access.

The reactions have been immediate and instructive. Motley Fool asked if the new product was “too good to be true.” Senator Elizabeth Warren sent a formal letter raising concerns about user financial data and regulatory oversight.

The crypto markets, meanwhile, briefly pumped Dogecoin on speculation about crypto integration, despite X Money being described as a fiat-only product with no confirmed crypto functionality.

Boosters are calling it the beginning of America’s super app moment. Skeptics are calling it a regulatory grenade with a debit card attached. Both camps have a point. Neither has the full picture.

X Money: The Product

First, the product itself. X Money offers a 6% annual percentage yield on deposits, a metal Visa debit card personalized with each user’s X handle, zero foreign transaction fees, 3% cash back on purchases, peer-to-peer payments, and FDIC-insured deposits held by Cross River Bank up to $250,000 per person.

X Payments LLC has received money transmitter licenses in 41 US states, enabling X to legally store, receive, and send money on behalf of its users. The rails are real. The question is who gets on the train—and why.

The Case For X Money

The lazy argument for X Money is distribution: tens of millions of US users, no new app to download, payments baked into a platform people already use daily. That argument isn’t wrong, it just ignores an important question: why would anyone change their payment behavior? There are some reasons:

• Broken creator payments. X currently pays creators via an ad revenue share program. But that money exits the platform immediately by going directly to a bank account. X Money shuts that door. The value isn’t in the transaction—it’s in the velocity of money staying inside the platform.

• Formalization of social transactions. Journalists, political commentators, podcasters—people who live on X professionally—already operate an informal micro-economy on the platform through Spaces, subscriptions, tips, and crowdfunded projects. For these users, the switching cost isn’t high because the behavior already exists—it just lacks the infrastructure.

• Social context. Venmo’s masterstroke was the social feed. X already generates the shared moments: breaking news, live sports, viral debates, real-time cultural events. The spontaneous payment behaviors those moments create don’t have a natural home on Venmo or Cash App because those platforms lack the social context that triggers the impulse. X has that context natively. X Money just needs to be present when the impulse fires.

• 6% interest. The target market isn’t sophisticated savers with accounts at Ally or SoFi, it’s 25 to 44 year-old X users who haven’t opened high-yield savings accounts yet. For them, an ad for a 6% yield appearing inside an app they open multiple times a day is a low-friction trigger. The switching cost is near zero because they haven’t committed anywhere else.

The Case Against X Money

The reasons why X Money won’t succeed outweigh the reasons why it could, however:

• The engaged user base isn’t as large as the headlines suggest. The creator economy and influencer class that represent X Money’s strongest use cases number in the hundreds of thousands—not millions.

• The platform has been shrinking, not growing. Mobile daily active users fell 15.2% year-over-year in 2025. A payments platform needs network effects moving in the right direction. X’s are moving the wrong way.

• Lack of trust. Venmo, Cash App, and Chime all built trust slowly, deliberately, and with a narrow initial use case. Musk has a history of jumping the gun with product announcements and timelines. That pattern matters when you’re asking people to deposit real money somewhere.

• Hostile regulatory environment. The timing of X Money’s 6% yield collides directly with Congress debating the CLARITY Act on yield-bearing products, and whether nonbanks should offer deposit-like returns. Senator Warren appeared to have her letter to Musk locked and loaded, just waiting for the product announcement.

• Platform toxicity. A meaningful segment of potential fintech users has disengaged from X entirely because of the platform’s political identity. You can’t separate X Money from X. No major payments platform has ever had to launch carrying that particular burden.

• The payments habit moat. Payments are habit-driven behaviors. People don’t switch payment apps because a competitor has better cash back. Venmo’s and Cash App’s installed user bases are large and their behaviors deeply ingrained. X Money is asking users to rewire behavior that has been reinforced thousands of times.

The X Money Super App Myth

Every X Money conversation eventually circles back to WeChat. Musk has been explicit about the aspiration: X as America’s WeChat. I wrote about this in 2022, and the argument hasn’t aged out.

WeChat isn’t a super app because Tencent had a clever product vision. It’s a super app because of conditions that are structurally absent in the United States.

WeChat’s rise was enabled by two specific factors: unbanked populations in China jumping to mobile payments, and strict app marketplace regulations that prevented Google Play and a fully open Apple App Store from operating.

As Insider Intelligence noted, “to plug the gaps, apps like WeChat and Alipay allowed third-party developers to integrate mini programs for other services, transforming them into super apps.” Neither of those conditions exists here.

What does exist in the US is something WeChat never had to contend with: a fully banked population with deeply entrenched app habits, a wide-open app marketplace, data privacy concerns, and an economic structure dominated by oligopolies in every major industry—oligopolies that will resist any platform trying to absorb their customer relationships into a walled garden.

There’s also a fundamental misread of American consumer behavior driving the super app hype.

The “hankering for service rebundling” isn’t coming from consumers—it’s coming from providers like PayPal and fintechs who see their narrow product lines as barriers to growth.

Americans have never wanted a one-stop shop in financial services or retail. The walled garden approach is too limiting for most US consumers.

If anything, Americans want the opposite: financial services embedded in the non-financial apps they already use—Home Depot, PlayStation, their favorite retailer—not a social media platform trying to be their bank, their broker, and Venmo simultaneously.

In addition, there’s a technical reality that gets overlooked: true super apps don’t work by simply piling features onto a mobile app. They rely on mini programs—lightweight apps that run inside a host app without requiring separate downloads or app store approvals.

That architecture is what made WeChat’s ecosystem coherent not bloated. X isn’t building that. It’s building a feature stack. Those are very different things—and the difference matters enormously for whether the user experience holds up at scale.

The super app framing isn’t just strategically misguided for X. It’s a distraction from the more achievable—and more honest—version of what X Money could actually become.

The X Money Prediction

X Money won’t become the super-app payments layer Musk envisions. “To be the place where all the money is, the central source of all monetary transactions” is classic Musk hyperbole that deserves the raised eyebrow it’s getting.

But dismissing X Money entirely may be premature. It could find traction with two segments if X executes on product fundamentals rather than press releases:

1. X creators. Content creators, journalists, podcasters, and independent commentators who already earn on X have a concrete reason to keep money inside the platform. Watch creator transaction volume, not total account openings, as the signal that X Money is working.

2. Yield-motivated procrastinators. There are millions of Americans who have been meaning to move money into a high-yield account but haven’t. They’ll put a few thousand dollars in which could be enough to build a real deposit base. The risk, though, is regulatory and sustainability: if the yield gets cut or capped, this segment evaporates as quickly as it arrived.

Neither of these rises to a WeChat story or anything that threatens PayPal.

The path to X Money’s success runs through the users the platform already serves well. The game plan should be: 1) build the closed-loop creator economy first; 2) use the high yield to attract the procrastinators; and 3) prove the product is trustworthy with real money before declaring victory over the competition.

A Harvard report claims that credit card interchange fees result in an annual transfer of $9.2 billion from lower- to higher-income consumers, with cash and debit users subsidizing credit card rewards. The math doesn’t hold up to scrutiny.

Credit Card Rewards Don’t Move Money Between Consumers

The paper claims that $9.2 billion “flows” from households earning under $150,000 to those earning more. That framing implies money moving from one consumer to another. It doesn’t work that way.

Everyone pays the same price at checkout (well, we used to). Some get some money (or points) back from their issuer. That’s not a transfer between consumers—it’s a difference in how much value each person extracts from the payment system.

The “transfer” framing requires the following causal chain to hold: 1) interchange fees raise retail prices above what they would otherwise be; 2) higher prices harm cash users; 3) therefore cash users subsidize rewards.

The chain, however, doesn’t hold for three reasons: 1) interchange doesn’t have that big an effect on prices; 2) credit card rewards can’t be accurately measured; and 3) spending on premium credit card reward cards occur in categories where lower-income consumers are absent.

1) Interchange Has Minimal Impact on Prices

The Durbin Amendment cut debit interchange, reducing fees on debit cards issued by large banks. If interchange fees drive retail prices, then retail prices should have dropped when interchange fell.

They didn’t. Prices kept rising after Durbin, as they had before it, driven by labor costs, supply chains, energy prices, and all the other factors that move prices.

No study has documented a meaningful retail price reduction attributable to the Durbin Amendment. Merchants absorbed the savings, or more precisely, the savings dissolved into the general cost structure that produces retail prices.

If lower interchange doesn’t produce lower prices, then higher interchange doesn’t produce higher prices. And if higher interchange doesn’t produce higher prices, the mechanism that supposedly creates the “transfer” doesn’t exist.

2) Credit Card Rewards Can’t Be Accurately Measured

Even if you set aside the price mechanism problem, the paper’s framework has a second flaw: it can’t reliably measure what premium card users actually receive in return for the interchange their spending generates.

The paper’s approach requires a precise measurement of how much value flows back to each payment type as a result of interchange fees. For cash back cards, it’s a simple calculation—a 2% cash back card returns 2 cents per dollar.

For points-based rewards programs that dominate the premium segment, it’s different. The value of a point can’t be determined until redemption—and is highly variable depending on how it’s redeemed.

A Chase Ultimate Rewards point redeemed for a statement credit is worth roughly 1 cent. The same point transferred to a airline partner and redeemed for a business class seat might be worth 2 cents or more. A reward point redeemed for a gift card at an unfavorable rate might be worth 0.6 cents.

The Harvard paper treats rewards as a uniform percentage return, ignoring a variance that can exceed 300%.

Then there’s breakage. A meaningful percentage of points are never redeemed—they expire, accumulate in dormant accounts, or are forfeited when cardholders switch products.

That unredeemed value represents pure profit for issuers. If a substantial share of premium card rewards are never collected, the actual value returned to consumers is materially lower than the interchange revenue that funded them.

This isn’t a rounding error. If those rewards are worth less than assumed, or if a significant share are never redeemed, the disparity in value capture between premium card users and cash users shrinks. The $9.2 billion figure becomes even harder to defend.

3) Credit Card Rewards Spending Happens Where Lower-Income Consumers Are Absent

The paper’s redistribution calculations acknowledge something the authors called “overlap:” consumers using different payment methods shopping at the same merchants.

But their analysis understates how much premium card spending happens in consumption categories—like business class airfare, luxury hotels, fine dining, high-end jewelry—where lower-income consumers simply aren’t present.

A cash-paying consumer can’t be affected by the interchange fees at a merchant they never patronize. No overlap means no redistribution, full stop.

The paper’s own data hints at this. The travel sector shows 70% credit card share and minimal cash presence, and the authors note this limits redistribution there.

But the logic isn’t followed to its conclusion: if a substantial share of premium card volume is generated in luxury categories that are inaccessible to lower-income consumers, the redistribution estimate is overstated not just by sorting within shared merchant categories but by the absence of lower-income consumers from entire spending categories.

If you’re not buying business class plane tickets or staying at five-star hotels, you’re not exposed to whatever price premiums interchange might create in those categories.

Result: Lower-income consumers can’t be harmed by the impact of interchange on prices at merchants they never visit.

Merchants are Addressing the Credit Card Rewards Issue

There’s a fourth development the paper doesn’t account for: the growing prevalence of card surcharges at the point of sale.

Recently, merchants have added explicit surcharges for credit card transactions—isolating the interchange cost and charging it directly to the consumer generating it, rather than baking it into uniform prices for everyone. Cash users pay the base price, credit card users pay a surcharge.

This is the opposite of the mechanism the Harvard paper assumes. The authors’ redistribution story requires merchants to apply interchange costs across all consumers through uniform pricing. Surcharging does the reverse—it assigns the cost to the specific consumer whose payment method creates it.

Surcharging isn’t universal, for sure. Many large merchants don’t impose it, held back by competitive pressure, customer friction, or contractual arrangements. So the trend doesn’t collapse the paper’s argument entirely.

But it does do something the report’s policy recommendations don’t: it aligns costs with the consumers generating them, without requiring an interchange cap, without eliminating rewards programs that millions of consumers value, and without repeating the Durbin experiment that failed to help the people it was designed to help (except, of course, the merchants who pocketed the difference).

Which raises the question the report never asks: if merchant surcharging already moves the system toward the outcome the authors imply policy should engineer, what problem does a legislative intervention actually solve?

Why the Credit Card Rewards Framing Matters for Policy

The paper’s “transfer” framing is already circulating in policy circles. The $9.2 billion figure will appear in Congressional testimony and advocacy groups will be sure to cite it. Proposed remedies will follow: cap interchange, restrict premium card rewards, extend Durbin-style regulation to credit cards.

The problem is that Durbin already ran that experiment. Interchange fell. Prices didn’t follow. The consumers the regulation was designed to help—lower-income cash and debit users—saw no meaningful benefit at the register. Middle-income debit users lost free checking. High-income credit users came out ahead.

Policy built on the transfer framing will produce interventions that won’t deliver the promised benefit. Why not? Because the mechanism that would deliver it—interchange rate cuts flowing through to lower retail prices—doesn’t work that way.

Capping interchange to reduce rewards that consumers may never actually collect is a poor trade for the middle-income households who would bear the cost through higher banking fees—as Durbin demonstrated.

The honest policy question the paper raises is narrower: should the rewards ecosystem be structured in a way that returns more value to lower-income consumers who participate in the payment system?

That’s a legitimate question. It doesn’t require a $9.2 billion transfer narrative to motivate it, and it points toward different solutions than interchange caps.

The Bottom Line on Credit Card Reward Policy

The Harvard paper demonstrates a possible disparity in value capture, not a transfer between consumers.

That distinction matters because the policy that follows from “lower-income consumers get less value from the payment system” is very different from the policy that follows from “lower-income consumers are robbed by affluent consumers.”

The credit card rewards system raises legitimate equity questions. The Harvard paper provides impressive data for examining them. It would be more useful if the framing matched what the data actually shows.