"What does Gen Z want from a bank?" is usually answered with adjectives. Authentic. Frictionless. Values-led. None of that means anything on the product roadmap, because you cannot build an adjective. So here is the question in a form you can actually build against: when you look at how under-25s spend, what they say in surveys, and the features banks are shipping to reach them, which expectations keep showing up?
Gen Z expects a bank to show them what they already pay for, to answer questions without making them dig, to reflect their spending back to them accurately, and to feel like an institution built with people their age in the room. Each one already has a feature attached. And underneath all four sits the same thing: a clean, enriched transaction record.
See what I already pay for
Start with a familiar small disaster: the subscription you forgot you had. You spot it three months late, on a bank statement, under a merchant name you don't recognise. Multiply that across a generation and you have the single widest gap between what Gen Z spends and what their bank can explain.
Visa put numbers on it last year. In research across more than four thousand UK adults, Gen Z came out as the heaviest subscribers by some distance, averaging £305 a month against £261 for millennials and well under half that for older users. Nearly all of them hold at least one subscription, which makes this group the most exposed.
On every failure Visa measured, from paying for something unwanted to missing a cancellation window to a bad experience with a Direct Debit, Gen Z's numbers ran about double the wider population's.In France, younger adults spend more than the national average and forget to cancel more often, with most having dropped or considered dropping a subscription for money reasons.
Our own reading of card transactions across Central and Eastern Europe adds a market-level view. Across all cardholders, subscriptions come to around 7% of card spending, roughly €900 a year for an active subscriber. That figure sits deliberately apart from the survey numbers above, because it is a different kind of measurement: transaction-derived, all ages, no self-reporting. It is here to size the category in a European market, not to be lined up against a Gen Z percentage.
Whatever the market, the product requirement is identical. To turn a subscription into something a user can act on, recurring payment detection has to produce three things: the frequency, the next billing date, and a merchant name the user actually recognises. Miss any one of those and the feature stops being useful: a "next payment" with no date, or a date attached to a descriptor the user has never seen, solves nothing.
Answer me without making me search
The second expectation is easiest to see through a contradiction Gen Z carries: they reach for AI constantly, and almost never trust it to act alone.
TD Bank caught this in a US survey this year. More than three-quarters of Gen Z said they'd used AI to help make a financial decision, the highest share of any generation. Ask the same people whether AI should make the recommendation on its own and only a small minority would hand it that. Deloitte finds the matching pattern on the other side of the interaction: younger customers report better experiences with banking chatbots than older ones do, and still want a human for anything routine.
The European version of the same shape comes from Česká spořitelna, Erste Group's Czech subsidiary and the largest bank in the country, with more than three million people using its George app. Its own research turns up something counterintuitive: the youngest clients value reaching a real banker more, not less. As the bank's Filip Hrubý has put it: “they know exactly what the app is good for and exactly where it stops, and they want a person waiting at that edge”. Fluency with the app and demand for a human don't pull against each other here. They arrive together.
There is also a hard legal line under any AI assistant a European bank builds, and it moved recently. From 2 August 2026, the EU AI Act requires an assistant to tell users they're talking to AI. A related obligation on AI-generated content follows at the end of that year. The heavier regime, the high-risk rules that cover things like creditworthiness assessment, was pushed back to December 2027 by the Digital Omnibus, which entered into force at the end of July 2026. If you've read anywhere that the high-risk clock started this August, that guidance predates the change.
Underneath the disclosure rules sits a simpler mechanism. An assistant can only answer from what the transaction record contains. Ask it why last month's spending jumped and a record full of gateway descriptors and bare MCC codes gives it nothing to work with. The quality of the answer is capped by the quality of the data, long before any model is involved.
Show me myself
The third expectation is for the bank to reflect the user's own behaviour back to them, accurately enough that they trust it.
Take the "year in review" recap, now a fixture in banking apps. It is worth treating as product reasoning: a twelve-month recap is only publishable if twelve months of records are clean. A single blank or missed merchant in that timeline undermines the whole piece, because the user notices the one entry they cannot place and stops trusting the rest.
Carbon insight is the same reasoning applied to sustainability. Kateřina Linhartová at Tapix makes the point that framing a lunch as "15 km by car" lands with a user where "3.5 kg of CO2" does not: the unit has to be one a person can picture. That framing only works if the underlying carbon coefficients are accurate, which for restaurants comes from price banding and for fuel from pump-level pricing.
Be an institution that looks like me
Erste Group runs full banks across seven countries, from Austria and Czechia through to Croatia and Serbia, serves well over twenty million customers, sits under direct ECB supervision as a systemically significant institution, and runs George as one shared digital platform across all of it, with more than eleven million users. Česká spořitelna is its Czech arm and the biggest bank in the country. That scale is the whole reason the next bit matters.
Its Future Mindset Board took six young people, chosen from nearly 2,500 applicants, paid them, and sat each one alongside a real board member with a brief to help rewrite the bank's strategy for the rest of the decade. The CEO's stated reason was refreshingly blunt: the board's average age was 54, and a group that old risks losing touch with the people it serves. One of the bank's leaders, Daniela Pešková, has framed the stakes with a single comparison, roughly a million and a half face-to-face banker conversations a year against close to a billion interactions with the app. When almost all your contact with customers runs through a screen, the people designing that screen had better understand who's on the other side.
There's a neat loop hiding in this. For this generation, the people you'd hire and the people you'd bank are the same people, so an institution that looks like it was built by twenty-somethings gets easier to work for and easier to sell to at the same time. The surveys can tell you the symptom; it takes someone inside the building to explain the mechanism.
Partners Banka shows the built-from-scratch version. It puts enriched data into the stack on day one rather than retrofitting it later, and even runs a financial-education pilot for children aged 9 to 15, meeting the users before they’re even a customer.
Four expectations, one dependency
All four expectations combine into a single technical problem.
A one-off tram fare and a monthly travel pass can land in your feed with exactly the same merchant description. Nothing in the name separates them; the only clues are the amount and whether it repeats. That's why working out what a payment actually is has to happen transaction by transaction, not merchant by merchant, and why it takes a genuine rule set behind the scenes, on the order of sixteen hundred subscription merchants and fifteen thousand-plus pricing rules, to tell a subscription from a lookalike.
Categorisation hits the same wall from the other direction. There's a single card code, MCC 4789, that covers taxis, buses, bike-share and e-scooters all together. A system that stops at the code sees one shallow transaction. One that goes underneath it, into dozens of categories and hundreds of store-level tags, can tell a bus ride from a scooter hire. That depth is the line between a feed that can power a feature and one that can't.
And the reason all of this bites for Gen Z specifically comes down to a shift in channel. The ECB's large euro-area study finds younger people making a markedly higher share of their everyday payments online than older people do, even as the card itself stays about as popular across every age group. They aren't paying with something strange. They're paying through a different window, and in that window the transaction record is the entire relationship.
That is the roadmap implication. Every expectation in this article terminates at the same enriched data layer, which is why the sensible next question looks like this: "is the data underneath clean enough to carry any of them." The feature-level companion to this piece takes the four expectations down to the individual features that answer them.
If you want to see what enriched output looks like on your own transactions, the Tapix sandbox runs it against your data.
