Last week I was at The Banking Scene conference in Brussels. A good chunk of European retail banking was in the room, and there was really only one topic: where the bank goes in a world where the customer increasingly talks to an interface, not to a branch.
I came home with a few takeaways worth unpacking. No marketing optimism, just what I actually saw and heard.
1. The agentic AI gap
Everyone talked about agentic AI. Yet nobody showed it. No demo, no live system, just architecture diagrams and roadmaps. That, in itself, is information.
My read is uncomfortably sober: the industry is significantly earlier than the conversations suggest. A slide with "autonomous agents that serve the customer end to end" looks great on stage. Between that slide and a working system in production - one a bank will let loose on real money and real customers - sits the whole of governance, security, auditability, and integration into core systems. That gap gets systematically underplayed at conferences.
The practical implication: if you do not have a single agent in production today, you are not behind. The whole industry is. The edge goes to whoever stops drawing diagrams and ships the first small thing that actually runs.
2. The System 1 problem nobody is solving
Noemi Molnar made a point that stuck with me. Most bank sales and marketing is aimed at System 2 - the rational, deliberate 10% of brain capacity. System 1, the remaining 90% that is emotional, subconscious, built on trust and relationships, is largely ignored.
As long as the customer sat across from a banker in a branch, a human handled System 1. But as banking moves deeper into conversational digital channels, that layer disappears - and it becomes the next frontier. Not just the right offer, the right channel, the right time. But the right framing, the right emotional register, the right form for each individual.
- System 2 is largely solved. Offer personalization, timing, channel - banks have put years into this and handle it reasonably well.
- System 1 is open territory. Tone, trust, the sense that the interface is on my side. Almost nobody does this in a digital channel.
- AI makes it possible at scale for the first time. Adapting not just what I say but how I say it, individually, for millions of people.
Not just the right offer at the right time. The right framing and the right emotional register for each individual person.
3. Revolut seems to actually get this
The company that seems to actually get this is Revolut. They build communities. They address System 1, not just the fee calculator.
On my way home through Brussels airport, I walked past a Revolut card vending machine - "Click. Collect. Spend. Free card here." - right next to a Starbucks. That is what System 1 acquisition looks like. No form, no waiting for the mail. An impulse, a gesture, a card in hand next to your coffee. Emotion and immediacy, not a product comparison table.
4. What Joris Krijger's data actually shows
Joris Krijger showed data I would put on the first slide of every internal AI presentation. Where a human is a genuine expert, the human still outperforms AI. But human plus AI outperforms both. And where a human is not an expert, AI alone is better - and adding a non-expert human to the loop actually makes things worse.
The takeaway runs against most of what gets said about AI: AI does not democratize expertise. It amplifies it. It gives a strong expert wings, it does not turn a weak one into a good one - and in some cases you are better off keeping them out of the decision entirely. That has hard consequences for where you deploy AI in a bank and who stays in the loop.
If you are working out where to start with AI in your bank so it does not become another roadmap in a drawer, get in touch. I am happy to look at where your biggest gap between talk and reality sits.
