Your bank is solving the wrong AI problem

The promise of AI in banking is real. But success frequently gets defined as technological progress, not the customer being better off.

Banking is being reshaped 

Artificial intelligence is changing what's possible in banking. It’s allowing firms to predict what a customer is about to need and act when the time is right. Routine tasks can disappear into the background and jobs that took a phone call and a fortnight can be solved in a tap.

In 2023, the Consumer Financial Protection Bureau (CFPB) found all ten of the largest commercial banks in the US had deployed chatbots. Many were still rule-based then: decision trees and keyword lookups - the first generation of digital, automated service. 

Fast forward to today: generative AI is challenging those rule-based systems. It goes well beyond chat, and into services customers rely on. That’s where AI’s promise is truly tested.

It works until it matters

So how’s the AI revolution going? Not quite as well as the momentum suggests.

In early 2026, a J.D. Power study found that among app customers, only 28% use virtual assistants - the AI people are most aware of in banking, and the one with the most data on how it’s working. Among those who do, satisfaction holds up on simple transactional tasks and falls away sharply when someone tries to resolve a problem, dispute a charge or report fraud.

The projects aren't faring much better. 100 retail and commercial banks were surveyed in early 2025 and, of the GenAI use cases those banks identified, only 16% reached full deployment. And of the ones that did, 40% failed to meet expectations.

Pressure is not evidence, technology is not a proposition

Competitive pressure, cost targets and business optics are legitimate concerns for any business. These are also, it seems, what drives many AI projects, and the evidence is accumulating.

A recent study by Boston Consulting Group shows that 61% of CEOs think that their boards are rushing AI initiatives, and over 50% say that the current hype is distorting judgement. 

Even more interesting is another detail from the same research that suggests that uncertainty translates into perceived urgency. It’s the board members who are the least confident in their own AI knowledge that are the most likely to think that their organisation is moving too slowly. 

The CFPB is fairly blunt about why banks started introducing automated customer service in the first place: “in large part to reduce the costs of human customer service agents”.

Notice what wasn’t mentioned? The customer, their experience or their money. A problem that needed solving. There’s little reason to think the AI that follows is driven any differently.

What pressure, cost targets and optics have in common is that they're satisfied the moment the new thing exists. Movement becomes the success metric, and having something to announce becomes the goal. At its core, nothing demands that the change is an improvement over the current state. 

A study by RAND, a non-profit research organisation, into why AI projects fail suggests where this takes us. In its interviews with data scientists and engineers, 84% pointed to leadership decisions and expectations as the primary reason projects fail. 

It's worth being clear about what that number is: product teams being asked why projects fail, and answering by pointing up. That’s ahead of data quality and tooling, and ahead of the limits of the technology itself. 

RAND also found that in failed projects, leaders often don't realise that "the metrics measuring the success of the AI model do not truly represent the metrics of success for its intended purpose".

So, the patterns are clear. Focus on the wrong problem, move on hype rather than evidence, and the project can hit its target and still miss the point. 

This is the problem. Technology alone can’t be a proposition. Customers don't choose banks because of technical architecture. They choose it because of what it offers them, and what it lets them do.‍

What it costs, and who pays the price

When AI is built without an understanding of customers, their situation, or how they will use it, wasted money is the smaller cost. The bigger one falls on customers, at the moments they most need the bank to work.

When it doesn’t, customers end up in doom loops, going round in circles with an assistant that can't help and that can't hand them to someone who can. Or they bring it problems the AI wasn’t built for, and get confident answers based on missing or out-of-date information. 

They may be told something that isn't true, or have their own mistake confirmed, be given financial advice that the bank would never have signed off, or get a different answer each time they ask. 

The service can change from one month to the next because the model underneath it was updated. And where AI makes decisions or takes actions, customers can be treated unfairly by a model that wasn’t properly tested, or find it doing something different from what they ask for.

Of these examples, doom loops are probably the best measured. In a 2026 survey of UK banking app users by payments firm DECTA, more than half of those with a blocked payment first tried to reach a person, but three in four ended up with a chatbot, which solved the problem 11.4% of the time.

So the customer pays for this, in time, fees, missed payments or wrong decisions, due to misleading information. 

Little of this may show up in the numbers a project is judged by internally. An assistant is often measured on containment rate, how many conversations it handles independently, other projects on their deflection rate, and a customer who gives up may count the same as one whose problem was solved.

Banks pay a price too. Clients’ loss of trust in their bank may not appear on the analytics dashboard, but it does affect what customers do next. After poor service experiences, Deloitte found that 28% of customers reduced their spending with their bank, and some left altogether. ‍

Real impact start with real problems

Yes, AI is making things possible that weren't just a few years ago. That makes it easy to think about AI as a kind of magic dust, something to sprinkle over an experience or an organisation to scale or improve service delivery, enhance productivity, or optimise operations. 

But it only works if we truly understand the context it lands in, and what a good outcome looks like for the customer. 

That matters more as systems become able to act. 

When an automated system can't help, the customer may get stuck, like the people we saw caught in doom loops. When the system can act on their behalf, a poor fit can move their money, make the wrong investment decision, point them to misleading information, or make commitments they didn’t want. 

With that much at stake, adding AI because it is available is a poor strategy. 

It may look like progress and it may answer internal pressure to move forward quickly, but that comes with clear, evidenced risk of producing products that fail at the point where people need them to work, eroding trust and impacting long-term retention. 

The solution is straightforward: start by understanding what customers do, how they think, where they struggle and what they are trying to achieve, then work out how to make the situation better.  

That’s how AI-native products start: with a customer problem, properly understood. 

‍

Now, let's bring it to life

At 11:FS, we don't just design digital propositions; we help you navigate the complexity of the growth cycle - from validating unmet customer needs to launching scalable future ready propositions.

If you are struggling to define what part AI plays in solving your customer problems, or how it fits within your existing experiences, let’s talk.

Find out more today at 11fs.com/ventures