The AI revolution in banking isn’t coming, it’s here. But if you talk to the AI companies actually selling into financial institutions, you will discover that widespread adoption is moving slower than the headlines suggest. Not because the technology doesn’t work, and certainly not because banks don’t need it. The roadblocks are more nuanced, more human, and more organizational than you might expect.
Over the past week, I sat down with five AI company founders who are deep in the trenches of selling to banks: John Sun from Spring Labs, Kareem Saleh from FairPlay, David Moscatelli from Abacus, Kalyani Ramadurgam from Kobalt Labs and Alec Crawford from AI Risk. What emerged from these conversations is a fascinating picture of an industry in transition, where the technical barriers have largely been solved, but the human and organizational ones remain stubbornly persistent.
The Universal “Not Now” Problem
Every AI company executive I spoke with had a remarkably similar experience: banks universally acknowledge they need AI, but getting from acknowledgment to implementation is where things get complicated.
“I don’t think we’ve ever gotten a ‘no, we don’t need this,'” John Sun from Spring Labs told me. “I think generally the rebuttals that we do get tend to be things like, ‘it’s not in budget for this quarter,’ or ‘it hasn’t risen to the top of our priority list, reach out next quarter.'”
Kalyani Ramadurgam from Kobalt Labs echoed this sentiment: “It’s never just a no. The challenge is, it’s often a postponing. They’ll say, ‘Hey, we can’t look at something like this until next month, we’re in the middle of an exam. We can’t get the budget until next quarter.'”
This isn’t the classic technology adoption curve where early adopters embrace innovation while laggards resist change. In banking, it is organizational inertia, along with strategic caution for anything new, that results in a slower adoption curve.
The Fear Factor: Data Security Trumps Everything
While banks may not say “we don’t need AI,” they consistently express two core concerns that shape every conversation: data security and accuracy.
Kalyani breaks it down: “The first piece of pushback we get really comes down to a hesitation of uploading their documents into a platform that will feed into an AI model. They always say, ‘what’s going to happen with our data?’ That’s almost always, without exception, the number one question we get asked.”
This data anxiety runs deeper than simple privacy concerns. Banks are acutely aware that regulatory penalties for data breaches can be devastating. As Alec Crawford from AI Risk explained, “Gramm Leach Bliley provides for a fine of $100,000 per incident for the bank, and $10,000 for the individual who’s responsible. And by the way, they get to decide how to define an incident.”
The solution, according to these AI companies, isn’t just technical, it’s about transparency and control. David Moscatelli from Abacus emphasizes their approach: “The client has complete control over the LLM. They get to choose what version of the LLM they’re operating on. They get to choose what data it’s connected to. They get to choose what data comes out of it.”
Then there is the challenge of vetting the different AI vendors for the bank, which sometimes feature AI agents doing semi-autonomous work. Kareem from FairPlay says, “Banks want some kind of solution for vetting the different AI systems and agents to understand if they actually do the things vendors claim. FairPlay’s platform provides this validation layer to allow banks to test these different systems.”
The Accuracy Imperative
The second major concern is accuracy, and here’s where AI companies have had to innovate beyond standard AI approaches. Banks can’t afford hallucinations when dealing with customer inquiries or compliance requirements.
The key insight is that banks don’t just need accurate AI, they need provable accuracy. Kalyani from Kobalt Labs puts it well: “Our AI will never say anything without giving you the exact explanation as to what the regulation says, highlight the internal policy and exactly where it has the issue, and then give you a very precise explanation about why it’s an issue, who’s gotten a citation before about this, and what you should do instead.”
This “trust but verify” approach has become standard practice. As David from Abacus noted, banks typically spend “three to six months before they really trust it” because initially, they make sure to verify every single answer.
The Regulatory Credibility Factor
One of the most important insights from these conversations is how crucial regulatory credibility has become for AI vendors. Banks aren’t just buying technology, they’re buying regulatory cover.
Kareem from FairPlay explained what banks really want to know: “Do we have credibility with their examiners and with their regulators? So that when they’re subject to supervision and they say, ‘hey, we rely on FairPlay for this testing, tuning and validation,’ the regulators will say, ‘yep, those guys know what they’re doing.'”
This has led successful AI companies to invest heavily in regulatory relationships and compliance expertise. FairPlay, for example, was selected by the New York Department of Financial Services as its advisor on algorithmic fairness issues, while Abacus is having its LLM audited by a top four accounting firm.
The Infrastructure Challenge
One of the most surprising insights from these conversations is that the biggest barrier isn’t always technical sophistication, it’s basic infrastructure readiness.
Alec from AI Risk described the reality: “The first roadblock tends to be technical, because in general, what we found is the technical expertise around AI but also the infrastructure required for AI is hard to integrate.”
This creates a chicken-and-egg problem: banks need AI software infrastructure to deploy AI tools, but they don’t have the expertise to build that infrastructure. Smart AI companies have responded by making implementation as turnkey as possible. David from Abacus boasts a “two-hour install guarantee” for their on-premise LLM solution.
The Bureaucracy Bottleneck
Perhaps the most frustrating challenge for AI companies is navigating bank bureaucracy. Even when there’s executive buy-in and budget approval, organizational complexity can derail implementations.
Kalyani from Kobalt Labs captured this perfectly: “Whenever we get in front of somebody who has the power to bypass bureaucracy, we almost always win those deals. It’s always about navigating the bureaucracy of a bank.”
The timeline implications are significant. Alec from AI Risk reports that “demo to sign contract is on average for us about seven months,” while some clients take up to two years to close.
The Education Gap
A consistent theme across these conversations was the need for extensive education. Banks may be mandated by their boards to adopt AI, but many still don’t understand what’s actually possible.
Kalyani from Kobalt Labs observed: “I’ve gone to so many conferences, and every single time I talk to these banks… I hear the same complaint, which is there’s so many AI panels where people are telling me AI is coming, it’s time to start planning for AI… But nobody is telling us what to do. They’re just telling us that it’s coming.”
The successful AI companies have positioned themselves as educators first, vendors second. Alec from AI Risk explained their approach: “We go over tons and tons of educational stuff. And then the last slide is like, oh, and we sell software to do this. It’s a very collaborative, educational, low-pressure approach.”
The Competitive Advantage Window
Despite these challenges, AI companies working with banks are seeing strong retention and expansion. John from Spring Labs noted, “The stickiness is great. We haven’t lost a production deal yet, and we rarely lose deals in proof of concept.”
Kalyani from Kobalt Labs reported similar results: “We have zero churn so far. So no bank customer of Kobalt has ever decided to stop using Kobalt or stop renewing their contract.”
John also talked about the ROI on implementing these new AI systems: “Our average deals are about 4 to 8x year one ROI, but can be deployed very, very quickly because it really sits on top of their existing tech stack.”
This stickiness creates a competitive moat. Once banks overcome their initial hesitation and successfully implement AI solutions, they become deeply embedded in daily workflows. The switching costs, both technical and organizational, become prohibitive.
The Path Forward
The AI companies that will succeed in banking are those that understand this isn’t just a technology sale, it’s an organizational transformation sale. They need to address data security concerns with technical solutions, accuracy concerns with transparent verification systems, and implementation concerns with turnkey infrastructure.
Most importantly, they need to position themselves as partners in navigating regulatory complexity, not just vendors selling software.
The window for early adoption advantage is still open, but it’s narrowing. Banks that embrace AI solutions now will have significant operational advantages over those that continue to postpone. For AI companies, the message is clear: the technology works, the business case is proven, and the regulatory environment is increasingly supportive. The challenge isn’t building better AI, it’s building better partnerships with the humans who need to trust it.
As David from Abacus put it, banks are dealing with “the illusion of progress with AI” through internal committees and planning sessions. “They’re trying to figure it out,” he explained. “They’re trying to move as fast as they can, but there’s a lot flying at them right now.”
The AI companies that can cut through that noise with practical solutions, regulatory credibility, and patient partnership will define the next generation of banking technology. Those who can’t will find themselves perpetually scheduling “next quarter” calls.
To learn more about AI implementations in banking, attend the AI-Native Banking & Fintech Conference on September 30 in Salt Lake City. Future Nexus is the exclusive media partner, so readers can receive 20% off their ticket by using the promo code NEXUS20.
