AI First
85% of banking users in Europe reject AI executing financial actions autonomously. But 80% want active smart analytics. The design challenge in banking is not how much AI — but where and how.
The Spanish banking user wants AI to help them decide. Not that I decide for him.
There is a recurring misunderstanding in AI projects for banking. It consists of assuming that, because users quickly adopt technology, they are willing to hand over control of their financial decisions to an automated system.
The data says otherwise.
A recent study by Sopra Steria with digital banking users in five European countries — including Spain — reveals that more than 85% of participants show active resistance to an AI system executing financial actions autonomously. Automatic payments, transfers, decisions that directly affect your money: AI does not rule there. They rule.
And yet, the same study shows that more than 80% of those same users are very interested in intelligent analysis features: spending summaries, automatic categorization, budget alerts, balance projections. They want AI active, useful and present. Just in its place.
The bank that already works well
There is a piece of contextual information that changes how all of the above should be read: almost three out of four users consider that their current banking application is simple and reliable. In markets like Spain, Norway or the Netherlands, the efficiency of banking apps is already perceived as the minimum expected standard — not as an advantage.
That means banking product teams aren't competing against a low bar. They are competing against an expectation formed by years of iterating on an experience that already works. Adding AI on top of that isn't automatically an improvement — it can be a source of friction if not designed well.
The risk of the race towards AI in banking is not falling behind technologically. It is building experiences that users do not understand, that they do not trust, and that they abandon in favor of the traditional interface that they continue to prefer.
The Spanish user: pragmatic and concrete
The study reveals clear differences between countries. Spanish users stand out for being especially pragmatic: they evaluate AI almost exclusively in terms of concrete usefulness and speed. They are not seduced by technological sophistication in itself — they are interested if it solves something real for them in less time.
That has a direct implication for design: in Spain, an AI function that does not have an immediately understandable use case is very unlikely to be adopted, no matter how powerful the model underneath is.
The question banking product teams should ask themselves before launching any AI feature is not “what can this model do?” but "at what specific point in the user's financial life does this really help them?"
Copilot, not autopilot
The metaphor that emerges most strongly from the study is that of the co-pilot. Users want a system that helps them understand what is happening, that alerts them when something deserves attention, that gives them context to make better decisions. But the decision is made by them.
More than 70% of participants say they need to understand why the system makes a recommendation before acting on it. An AI that does not explain its reasoning generates distrust — even when the bank that deploys it has a high level of institutional trust.
This has very specific design consequences. The moment of the recommendation — how it is presented, what context it accompanies, what options the user offers to intervene or reject it — is as important as the quality of the model that generates it. A good model with a bad recommendation interface produces distrust. An acceptable model with a transparent and controllable interface produces adoption.
The transition that is to come
More than 75% of users see AI as an additional layer on top of the banking interface they already use — not a replacement. Today, users prefer AI-augmented interfaces rather than AI-first interfaces where conversation with an agent replaces navigation.
That doesn't mean that moment won't come. It means it hasn't arrived yet. And organizations that force it before the user is ready will encounter resistance.
Those that do it well will be the ones that design that transition gradually: first adding analysis functions that build trust in the AI, then transparently offering suggestions, and only when that trust is built, expanding the scope of what the system can do autonomously.
Rushing into banking AI is not a competitive advantage. Trust built step by step, yes.
