Every bank has an AI strategy. Most have an AI mess. Not publicly, of course! The press releases are cheerful, the CIO slides show amazing adoption curves.
By: Vytenis Sakalas, Co-Founder & CEO at Moterra
But behind the polished narrative, the same conversation keeps happening in corridors and internal calls: the rollout that stalled, which LLM might be better, the AI vendors nobody trusts, etc. And still everybody dreams about perfect AI tool.
Where it goes wrong
Three of the most common causes for failures:
Lack of business context. A compliance officer runs a KYC check through a generic AI tool. The answer comes back structured, confident, cited. It’s also based on a regulatory framework outdated eighteen months ago. The AI answered from the internet, not from the institution. In regulated environments, a plausible-sounding wrong answer is more dangerous than an obvious one.
Shadow AI usage. Leadership announces the firm is evaluating AI options. Meanwhile, half the team is already pasting client data into personal ChatGPT accounts because it helps them work faster. Nobody sanctioned it, but it’s happening – and every paste is a potential GDPR violation. The policy said don’t. The personal curiosity and workload said do. The workload won.
Lack of trust – and this one hurts most because it already looked like progress. A Baltic payments firm did everything right on paper. Fifty ChatGPT Enterprise licences, board-approved. Reality: a handful of people used it. Nobody trusted putting real client information into a platform on someone else’s infrastructure. The licences were paid for, but the problem remained unsolved.
Last chance: Get your tickets for Baltic Fintech Days May 13
The advantage being built in silence
EY’s 2025 European Financial Services AI Pulse Survey found that 57% of financial services leaders believe their organisation’s approach to technology risk is already insufficient for the challenges AI is creating. Not in five years. Now. And yet the same firms are moving forward anyway, because the competitive pressure to do something outweighs the discomfort of doing it badly.
Huge banks and corporations have been solving this quietly – hundreds of engineers building private AI infrastructure from scratch. Proprietary knowledge layers built on their own data. It’s a structural advantage being assembled in the background.
Mid-size banks and fintechs don’t have the budget to build from scratch. They’re left with navigating between various AI tools in the market they could safely use on real data, or doing nothing.
What actually works
Baltic banks and fintechs have one big advantage the large banks don’t: speed. They can adopt the right architectures and tools in weeks, not years. The firms I’ve seen get real value from AI didn’t just move fastest. They moved smallest first. One workflow – a KYC check, a credit file review – run inside their own infrastructure where the model could actually access real business data. Then they built from there. No big bang deployment. Just one thing that genuinely made someone’s job easier, then the next.
Last chance: Get your tickets for Baltic Fintech Days May 13
I run an AI company, so yes – I have my bets in this game. But that’s exactly why I see these failures from close. The AI itself is not the hard part. Getting it to actually know your business, your policies, your clients, your risk framework – without handing all of that to a third party – that is.
The firms that solve this first will be harder to compete with. And if there’s one region positioned to move fast enough to matter, it’s this one.
