Transcript
Hey everybody. Welcome back to another episode of What’s Going On in Banking. I’m Ron Shevlin, chief research officer at Cornerstone Advisors, and I’m here, of course, with my colleague Stacey Bryant.
We have to get right into today’s discussion because at the top of the list is JPMorgan Chase’s decision to charge fintechs, data aggregators, or somebody in that ecosystem for customer-data access.
Not surprisingly, that touched off a firestorm. The fintech community has not reacted positively.
What bothers me is some of the exaggerated language around it. I’ve seen people call this a data-rights issue, say it will hold back innovation, or argue that it will harm the families the financial system is supposed to serve.
One fintech influencer wrote something like, “So now JPMorgan Chase is going to charge me to access my own data.”
No. JPMorgan Chase is not charging you to access your own data. That is not what is happening.
The larger argument is that charging for data access will suppress innovation. I’m not convinced. We’ve had regulation, pricing, fees, and other constraints in financial services for a century, and innovation has continued.
I also don’t think this is automatically a charge to consumers, so I’m skeptical of the claim that it inherently harms families.
There are legal questions, and I’ll be the first to admit I’m not a lawyer. I didn’t even try to read Dodd-Frank or Rule 1033 in full myself. I leaned on a couple of large language models and still didn’t get a clear answer.
So I reached out to a couple of lawyers I know. Both told me it was inconclusive whether Chase’s approach is clearly prohibited by statute or whether the primary restriction comes from the CFPB’s Rule 1033, which the agency, at least in its current form, has indicated it may not enforce or may revisit.
So there are a lot of legal questions.
But at the most basic level, there is still the question of whether JPMorgan Chase should be allowed to charge for providing this service.
Strategically, I think it’s a brilliant move. Chase has put a stake in the ground and forced everyone to confront the economics of the system.
If you’re a fintech on the other side, of course you don’t like it. And there’s still the question of whether the cost is ultimately passed through to customers by fintechs or data aggregators.
But is it really fair to say one company or industry must provide an expensive service and is not allowed to charge for it?
Let me use an analogy, and I know people are going to poke holes in it.
Stacey, can you see the envelope I’m holding up?
Yep.
This is my envelope. I bought it. I have a receipt. There’s no question about who owns the envelope.
Now I’m putting a $20 bill inside it, and I want to send it to you.
The envelope is mine. The money is mine. But we’re not in the same room, so how do I get it to you?
I could get in my car and drive from Boston to New Jersey, or wherever the hell you live. That would be inconvenient, and it would cost money in gas and tolls, but I would know it got there safely.
If I had to do that regularly, it would be absurdly inefficient.
So I could go to the United States Postal Service and have them deliver it. That would be more convenient, but guess what? They would charge me.
What? It’s my envelope. It’s my money. Why are they charging me?
Or I could use FedEx or UPS and pay even more for faster or more reliable delivery.
Again, it’s still my envelope and my money, but I’m paying someone for the infrastructure and service required to move it from one place to another.
That’s the part of the Chase argument that I think holds up.
JPMorgan Chase and other banks have built infrastructure to move financial data securely, quickly, and reliably. Why should they automatically be required to provide that infrastructure for free?
And if they are allowed to charge, why shouldn’t the market play some role in determining the price?
The postal service is regulated. FedEx and UPS largely set their own prices. If one provider gets too expensive, customers have alternatives.
Banking is similar in that consumers still have choices. There are thousands of banks and credit unions.
If Chase directly charged its own customers for data transfer and those customers hated the fee, they could choose another institution.
But that’s not really what is happening. In many cases, the data is moving to another provider that wants to use the customer relationship Chase originally serviced.
So why should Chase automatically be required to perform that work at no charge?
Stacey, where is my argument wrong? Forget the legal issue for a moment. I know I’m not qualified to definitively interpret the statute.
But the claims that this will destroy innovation, harm families, or fundamentally violate data ownership don’t make sense to me. Where am I going wrong?
I think you’re pointing to something Jeff Bezos has said for years: “Your margin is my opportunity.”
For startups and fintechs that rely heavily on API integrations and data aggregation, this creates a new expense line. Chase is effectively reaching into their margin.
When I’m on the road talking with vendors, especially fintech companies, I always ask how the business model works. Some of them get irritated because I’m essentially asking, “How are you making money on this?”
But the margins can be very strong, sometimes in the 60% to 70% range.
So I think it was inevitable that somebody, whether Jamie Dimon or another large bank, would eventually say, “If our infrastructure is helping enable your business, we want to get paid for it.”
It’s similar to your mailing example.
My grandmother used to send me birthday cards with $10 or $20 inside. She still had to pay for the stamp.
I recently had to send some things to my brother in Spain. UPS wanted an astronomical amount. I shopped around, went to the post office, paid less, and accepted that it would take a few extra days.
Peer-to-peer payments work the same way. Venmo may charge you for instant access to funds, or you can wait a couple of business days and get the money without that fee.
So I think this kind of charge was inevitable.
And once you start touching margins, people are going to complain.
Fintech companies are already using AI and automation to improve efficiency. If a bank adds another material cost to the business model, that pressure has to go somewhere.
So how do I think it plays out? If I’m a startup and data access is essential to my value proposition, I either absorb the cost, pass some of it through, or find another way to improve the economics.
I don’t think the industry has a magical fourth option.
I’ll put a stake in the ground and say there are probably two broad outcomes.
One is regulatory clarity.
If regulators step in and define what banks can charge, I think the result could look somewhat like interchange regulation. Banks may be allowed to charge, but the fee could be limited based on some estimate of the cost of providing the service.
I don’t necessarily agree with how regulators have calculated interchange costs in the past, but at least that approach would be internally consistent.
The problem is that a regulatory process could take 12 to 18 months, maybe longer. That leaves a lot of uncertainty in the meantime.
The second possibility is that we get no near-term regulatory clarity.
In that case, I don’t see much stopping Chase from testing the model unless a future administration, Congress, or court changes the environment.
That gives Chase time to begin implementing fees and negotiating with partners.
And I can’t help wondering whether this creates an opportunity for smaller banks and credit unions.
Many fintechs are not literally stealing a banking relationship. They may provide a narrow product or service that complements what the financial institution does.
A midsize bank could potentially say, “We’ll provide the data at little or no cost, but we want something in return.”
Maybe that’s aggregated, anonymized information about customer behavior. Maybe it’s a distribution partnership. Maybe it’s something else entirely.
I suspect Chase itself may ultimately negotiate down from whatever opening position it takes in exchange for strategic value.
They’re not stupid. They know how to negotiate.
You make a good point about Rule 1033.
Under that framework, fees were supposed to be reasonable, not structured to undermine access, and not discriminatory or anticompetitive.
A lot of the fintech criticism seems to be leaning heavily on that third point. They’re arguing that the fee is effectively anticompetitive because it raises the cost of competing with large banks.
But at some level, if you want to do business using another institution’s infrastructure and customer data connections, you’re going to have to absorb some cost.
The fintech companies are kicking and screaming because their economics are being challenged.
I also think this could remain unresolved for most of the current administration, maybe even longer.
And remember, the direct customer of Chase may not even be the fintech. Often the data aggregator is the intermediary.
So another big question is whether the aggregator absorbs the fee, passes it on to the fintech, or restructures the business model.
There are a lot of moving parts.
Another topic we wanted to get into today is how large language models are being used inside financial institutions.
You saw an American Banker article about Goldman Sachs. Tell us about that.
Yes. The headline was something like, “Goldman Sachs Staff Now Write a Million GenAI Prompts a Month.”
It’s a great headline.
Goldman initially rolled out GitHub Copilot to roughly 12,000 developers. Now it’s giving developers access to Cognition’s Devin, an agentic-AI coding tool that can act more like a team member and execute multi-step tasks with limited human intervention.
The broader strategy is to integrate generative AI across the firm.
Goldman’s CIO, Marco Argenti, emphasized something we talk about constantly: there is no high-quality AI without high-quality data.
That should sound familiar.
Data strategy, governance, and warehousing are foundational because the institution needs confidence in the data before it can rely on the output.
Goldman also identified specific use cases that it believed added real value, addressed cybersecurity and data-protection requirements, and worked to reduce hallucinations before expanding access.
Then it rolled the tools out across different levels of the organization, from analysts and developers to executives, marketing teams, and other employees.
Today the firm is reportedly seeing more than a million prompts per month, and usage is still growing.
What I like is that Goldman says it is working backward from internal use. Start with employees. Improve productivity in areas such as client presentations, financial analysis, internal communication, and software development. Let employees share what works, then scale the strongest use cases.
The firm is also comparing models such as Gemini, Claude, and GPT to figure out which produces the most accurate output at the lowest cost.
The CIO even talked about using a virtual colleague and delegating work to it.
And then there’s the ROI question.
Goldman said it is measuring development productivity and translating some of that into cost savings. It expects agents to increase those savings further, but the bank admitted it does not yet know by how much.
That’s the part I like.
They’re not pretending the answer is fully known. They’re testing, learning, gathering employee feedback, and expanding what works.
You spend a lot of time with boards. How do you think this kind of approach translates to community and midsize financial institutions?
Let me address the article itself first.
I don’t remember the author, but I give them credit for trying to get something concrete out of Goldman’s CIO.
That said, the CIO managed to speak at length without giving us nearly as much useful detail as I would have liked.
A million prompts sounds impressive, but it’s a little like saying 17 billion hamburgers were sold. Were they good hamburgers?
What were the prompts actually about?
Was there any categorization?
How many were useful? How many were retries? How many came from people who didn’t know how to do something and asked an LLM to figure it out? How many simply replaced a Google search?
I use ChatGPT constantly. Sometimes I ask a question, don’t get what I want, and rephrase it four times. That’s five prompts to get one useful outcome.
So prompt volume by itself is not a meaningful productivity metric.
And this is why I think the ROI question is so often misguided.
Forty or 45 years ago, when PCs appeared on employees’ desks, economists complained that they couldn’t measure the productivity improvement.
Of course they couldn’t. Companies weren’t necessarily firing people. PCs increased capacity. They allowed the same employee to do more.
That is largely what we’re seeing now.
Unless Goldman is reducing headcount or avoiding hiring, it’s difficult to translate every productivity gain into a clean financial ROI.
The CIO also sidestepped the question of which models were actually better. He talked about differences in reasoning capability, but we didn’t get a lot of actionable detail.
So I’m not sure the article is especially helpful for a midsize institution trying to decide what to do next.
If the takeaway is simply, “Turn on Gemini, Claude, and ChatGPT and let everybody use them,” that is not a strategy.
Institutions need much more focus at the use-case level.
Where are employees spending unnecessary time? Where can response times improve? Where can better information create measurable business value?
We recently did research at Cornerstone on the use of large language models for policies and procedures.
The banks we interviewed could not precisely quantify the financial return, but senior executives consistently said the time savings were significant.
In the old world, an employee might submit a policy question to somebody in compliance or operations and wait three days for an answer.
With an LLM trained on approved internal policies and procedures, they can potentially get that answer in seconds.
That’s a clear use case.
So I’m much more interested in the quality of those use cases than in whether a bank can brag about writing a million prompts.
I get that.
For me, the moral of the Goldman story isn’t that the million prompts are automatically impressive. It’s that one of the biggest institutions in financial services is still testing.
They’re not pretending every question has been answered.
They’re experimenting with tools, benchmarking models, gathering feedback from employees, and deciding what works.
The article even talked about holding AI systems to a level of scrutiny comparable to human workers, which raises a whole other set of questions. You can discipline an employee for a mistake. You can’t exactly reprimand a hallucinating model.
But the larger point is that testing plants the seed.
Think about animation. It started with pen and paper and eventually evolved into sophisticated 3D animation. The later capabilities would not exist without the earlier experimentation.
I saw the same dynamic recently at the African-American Credit Union Coalition conference in Atlanta. There was a standing-room-only breakout session on AI and technology.
A lot of the material was basic, but that was actually important. Banks and credit unions have multiple generations of employees and executives. Sometimes you have to start small and make the technology feel understandable before the organization can become more agile with it.
That was my biggest takeaway from the Goldman story: AI adoption can start with something as simple as helping an employee build a PowerPoint, and then grow from there.
Great point, Stacey. I’ll give you the last word on that one.
I think we’ll wrap up this episode here. Thanks for a great conversation.
And thanks to everybody for listening. We hope you’ll join us for the next episode.
If you enjoyed today’s What’s Going On in Banking episode, hit the follow button on Spotify, Apple Podcasts, YouTube, or wherever you’re listening. We’ve got more conversations coming your way, and you won’t want to miss them.
Thanks for tuning in.
Enjoying What's Going On In Banking?
Subscribe on your favorite platform