AI is transforming finance, but new infrastructure is needed for bots to take the lead
In the 19th century, the global ice trade was booming. Blocks of frozen water were shipped from New England to destinations as far as India and Australia. However, as fintech expert David Birch reminded the audience at Money20/20 Europe last month, the entire industry collapsed. Not because ice got easier to ship, but because consumers got refrigerators.
He described how those in the ice business initially welcomed refrigeration, believing it would help them transport ice more efficiently over longer distances.
“Completely missing the obvious point, which was that Big Tech was actually putting refrigerators in the customers’ houses,” he said.
“They weren’t in charge of their own destiny, because actually what they were doing was being replaced by the customers. And this is where we are with AI.”
Just a few weeks later, on July 15, AI research company Anthropic unveiled its financial analysis solution, positioning its Claude 4 model as a full-stack AI assistant for institutional finance.
The platform aims to unify market and internal data, integrate with providers like Snowflake, FactSet, PitchBook and S&P Global, and include prebuilt connectors, implementation support and Claude’s advanced code-generation capabilities.
Claude, according to Anthropic, passed 5 out of 7 levels of the Financial Modelling World Cup competition, an annual Excel modelling competition, and scored 83% accuracy on complex Excel tasks.
Institutions are already using it in production. “Claude powered the first versions of our Investment Analyst Assistant,” said Aaron Linsky, CTO at AIA Labs at Bridgewater.
“With Claude, we estimate that we have achieved 20% productivity gains – equivalent to 213,000 hours,” added Nicolai Tangen, CEO of NBIM.
However, a month earlier, Birch argued these kinds of productivity gains miss the deeper shift AI is bringing to finance and payments.
“This is why most of the stuff we hear about AI out here is tangential and irrelevant. Do you think it’s really going to reshape the future of banking if you can have 10% fewer people in the call centre?”
“It doesn’t make any difference. The future isn’t being shaped by what banks want to do with AI, what payments companies want to do with AI. It’s being shaped by what consumers will do with AI.”
“You’re deluded”: Birch warns banks are missing the point
Another twist in the prediction by Birch is that increasingly, those “consumers” may not be human.
“In the future, it’s not about you using AI to serve your customers better, because those customers are going to be AIs themselves,” he said. “You honestly think payments are so interesting, people want to do it themselves. You’re deluded. You work in one of those boring industries.”
Birch’s co-panellist, Interac’s Debbie Gamble, built on the point.“What AI is really about is a shift change in our business, a shift change in how we look at our customers, and indeed, how our customers look at us,” she noted
Gamble referenced Frederic Tudor, the so-called ‘Ice King’, whose business transporting ice was ultimately killed by the very technology he helped usher in.
“At first they thought, oh this is good, we’ll get to move the ice around faster, it’s more economical… until in fact what happened with that piece of technology is it completely eliminated his business.”
Building Infrastructure for bots, not just humans
For Gamble and Birch, this lesson applies directly to today’s payments infrastructure, especially with respect to the rise of agentic commerce, where bots act on behalf of users.
“We built this fantastic infrastructure… But what we believe is that AI is actually going to change fundamentally, not just that infrastructure, but a world where the bots are making the decisions on behalf of the customers.”
However, enabling bots to act safely and autonomously in financial systems isn’t just a matter of plugging them into existing rails. Birch and Gamble argue a completely different kind of digital public infrastructure is needed, one designed for non-human agents.
“We need a completely different set of policies, rules, frameworks for engagement that means that the importance of digital public infrastructure is hugely significant to this,” Gamble said.
This includes things like digital identity for bots, smart wallets with embedded policy logic and authentication protocols which don’t rely on a person pressing “confirm”.
“We don’t want AIs pretending to be people. We want AIs being AIs,” said Birch.
As more decisions are delegated to these agents, from investment choices to routine payments, new security, compliance and liability frameworks will become essential.
“Your next customer is most likely going to be a bot,” Gamble warned.
