What’s actually slowing down AI adoption in financial services? According to the experts, it’s not regulation

Discussions about AI in financial services keep circling back to the same question: what’s actually slowing adoption down? The instinctive answer is regulation, an easy villain: strict, slow-moving and safely outside anyone’s own control.

Four industry experts see it almost the opposite way. Jukka Moilanen from OP Retail Banking, Shyam Mohan from Backbase, Juha Raumolin from Banksmith and Claudio Sangiorgi from CRIF discussed the topic at the AI in finance event, which was built on Qvik’s report on the views of Nordic financial leaders.

Regulation isn’t the biggest barrier

According to Qvik’s report, regulatory requirements themselves aren’t the main obstacle. The bigger bottleneck is internal compliance checks, repeated separately for every use case and, especially in larger organisations, prone to dragging on. The fastest-moving organisations bring the compliance perspective in at the design stage, rather than leaving it as a final check before launch.

The panel’s own examples bear this out. Raumolin describes AI that transcribes investment advisor meetings and automatically flags compliance risks, such as whether a client’s ESG preferences have been properly considered: the rule wasn’t a constraint but the reason the tool got built in the first place. Sangiorgi recognises the same pattern in CRIF’s own work. “AI is not a magic hat,” he sums up: you first need a genuine business case and a clear boundary on what data the AI is allowed to touch.

According to Moilanen, what really holds things up is often how organisations interpret regulation themselves; it isn’t a lack of permission but organisational slowness. Mohan sees it more as a question of courage: he points to a multinational bank that lifted its sales performance by 20% within a year by daring to act boldly within the rules rather than waiting for certainty that never really arrives. “The people who dare to be slightly bolder are also the ones who can show real metrics and real impact,” he says.

Transparency alone isn’t enough

The EU AI Act’s requirement to tell customers when they’re dealing with AI sounds simple on paper. In practice, what matters most isn’t the disclosure itself but whether the AI actually resolves the customer’s problem, and what impression that leaves. The worst way to lose trust is to leave a customer stuck in endless, useless loops with a chatbot that goes nowhere.

What transparency means also depends on where you start from. CRIF has been building credit scoring models for 40 years, and for them, Sangiorgi says, transparency has always meant being able to explain a result on request, not publishing the model in full.

For OP, it’s above all a question of values: “It would go against our values not to tell our customers that we use AI in our services,” Moilanen says. That doesn’t mean generative AI belongs everywhere, though; a simpler intent-based model is often the safer choice.

ROI is real, but it’s not evenly distributed

The technology is the same for everyone, but what gets measured, and how, varies from one organisation to the next.

According to Sangiorgi, time savings of 70 to 90% in document processing are real, but some applications change the way work gets done so fundamentally that it stops making sense to measure ROI against the old process at all.

“You can’t just plug AI onto a bad process and bad data: it’s still a bad process and bad data,” Raumolin says. The real gains, he argues, come from systems built with AI as the starting point.

Mohan points out that average handling times have actually gone up rather than down, because customers increasingly choose to talk to AI, which eats into the efficiency gain. In his consulting work he has seen plenty of banks talk about 30 to 40% efficiency without yet having made the decisive call that would turn that figure into a real saving: cutting headcount, or using the freed-up capacity to absorb growth without hiring more people. Until one of those happens, the efficiency figure is a claim, not a measured result.

Customers are splitting into two groups

Some customers, shaped by ChatGPT and similar tools, expect their bank to match that experience directly; that isn’t something a bank can offer, not least because of regulation and data constraints, Moilanen says. At the same time, a vocal minority don’t want AI anywhere near their finances at all. Both groups are a minority: most customers simply want their mortgage processed faster and don’t much care whether AI is what makes that possible.

The same split runs across generations too. For now, the answer is to keep both paths open: a faster AI-assisted route and a slower, more human one, for as long as customers actually want the latter.

The value isn’t only in the product

Where attention gets directed also decides where the value turns up. Public discussion tends to focus on what the customer sees, but most of the actual work, and probably more of the value, happens somewhere the customer never looks.

Qvik’s report confirms this: right now, the most concrete and easily measured benefit sits in back-office processes, things like shorter processing times and cost savings in claims handling and dispute resolution, while predictive services and personalised offers remain longer-term goals.

According to Mohan, AI is moving from an assisting chatbot to something that resolves queries outright, and on towards a role orchestrating back-office processes themselves, such as raising a card dispute directly with the card scheme without a human in the loop. Moilanen gently pushes back on the whole “product” frame: for OP, the biggest value comes from redesigning processes, and from what he calls becoming an “AI-native organisation”: not what AI can do, but how roles and ways of working need to change around it.

Customer experience could disappear altogether

What happens when a customer’s own AI agent and a financial institution’s agent handle a matter between themselves, with neither side directly present any more? “As the world moves toward a situation where the customer’s agent and our agent handle the process between themselves, I’m somewhat concerned that no customer experience gets created in that scenario at all,” reflects a bank’s head of technology development interviewed for Qvik’s report.

This isn’t just a future scenario any more. Raumolin points to a Swedish online broker that has already opened up an interface giving customers access to their own investment data, and asks which Finnish bank will be first to let a customer get hold of their own data and choose the interface they use to work with it, powered by AI. The technology already exists. Whether a Finnish bank offers its customers the same option or leaves it to competitors is purely a business decision.

Six takeaways from the discussion

• How organisations interpret regulation slows AI adoption more than regulation itself.
• Transparency works when the AI also genuinely solves the customer’s problem.
• Measuring ROI is an organisational question: efficiency figures only become real savings once an organisation makes the decision to realise them.
• Most customers don’t care about AI, as long as things get handled quickly.
• Right now, the most concrete and measurable value is being created in back-office processes.
• Agent-to-agent transactions are already technically possible, and competitors may get there before your own organisation does.