The signal
Let's start with the strangest part.
Paper is back. Driven by social media creators, part of Gen Z has returned to cash: cash stuffing means physically dividing your banknotes into labelled envelopes — groceries, nights out, petrol — and stopping when the envelope is empty. Twenty-three per cent of young consumers say they use cash almost exclusively. The stated aim is not nostalgia: it is regaining control by making money visible and allocable.
The screen is winning. That same generation is twice as likely as average to hold a neobank as its primary institution — 10% against 6%. Seventy-six per cent prefer mobile banking, and two-thirds make it their main channel. They research on social platforms and trust their peers more than institutions.
And now the bot. AI-generated traffic to US banking sites rose 1,200% over a six-month period. Sixty-two per cent of Gen Z and Millennials say they would trust non-bank AI platforms with financial services. A third of consumers expect at least 10% of their purchases to be AI-driven within a year, and 72% of merchants believe adoption will outpace their own readiness. Visa has published its Trusted Agent Protocol to secure agent-initiated payment; Microsoft and PayPal have launched Copilot Checkout, letting people search, decide and pay without leaving the assistant.
So: paper envelopes on one side, autonomous agents on the other. Good luck with the media plan.
What everyone is saying
That Gen Z is elusive and volatile — “switch happy”, as the literature puts it. That cash stuffing is a TikTok fad of no consequence. That AI agents are the real threat, because in removing friction they remove loyalty, leaving banks to compete on headline rate alone. McKinsey and Oliver Wyman converge on the point: the agent risks inserting itself between bank and customer, pushing the brand into the background.
All of that is correct. And nobody explains why both behaviours coexist in the same people.
What it actually changes
The explanation fits in one sentence: these are not two opposing trends, but one demand addressed in two different places.
Dividing banknotes into envelopes and delegating account comparison to an AI are the same gesture: taking back control in an area where you feel out of your depth. In one case you deliberately slow down in order to feel what you spend. In the other you speed up so as to stop enduring a decision you don't know how to make. The constant is not speed. It is control.
Yet the sector has spent fifteen years optimising a single variable: removing friction. Two clicks instead of six. Four-minute onboarding. Instant transfer. That was useful, and it is now a standard everyone meets — which makes it an asset that differentiates nobody.
Cash stuffing is the empirical, faintly humiliating proof that friction is not the enemy. These people carry the smoothest banking app ever designed in their pocket, and they go and withdraw notes to slide into paper. This is not a rejection of technology, it is a rejection of the painless. A purchase that costs nothing to execute stops being felt.
So we should stop opposing friction and fluidity, and start separating imposed friction from chosen friction. The first — one form too many, a document to resend, a call queue — should go, without debate. The second — a pause before an impulse purchase, a gesture that makes a decision physical, a ritual that makes a sum real — is precisely what customers rebuild on their own when the bank doesn't offer it. They call it envelopes.
And the AI agent in all this? It doesn't pose the problem people assume. The real risk is not that a bot compares better than the customer: it is that there is nothing left to compare. If a banking brand exists only through the smoothness of its journey, the agent mechanically reduces it to a row in a table — a rate, a fee, a term. It doesn't destroy the brand; it reveals there wasn't one.
The strategic consequence is almost cheering. Facing a customer who is sometimes a human with envelopes and sometimes an agent with an API, the only tenable position is the same: have something to say that doesn't reduce to a number. A way of treating people, a stance, a recognised competence. What banks long called “the relationship” and gradually replaced with onboarding.
Two behaviours, one demand
| The cash envelope | The AI agent | |
|---|---|---|
| What the customer does | Deliberately slows down in order to feel what they spend | Speeds up to stop enduring a decision they can't arbitrate |
| What it says about them | They don't feel in charge of their day-to-day spending | They don't feel competent to compare financial products |
| What banks usually conclude | A passing fad with no strategic weight | A technological threat to the customer relationship |
| What it actually means | Chosen friction has a value the sector stopped seeing | Without a distinctive proposition, the brand becomes a table row |
| The brand lever | Offer rituals that make a decision physical | Hold a defensible position that survives a comparison with no logo |
The question to put to your committee next week
If an AI agent compared our offer with our six competitors', in a table, with no logo, no colour and no campaign, would there be a single row where we win?
If yes, that is where to invest, and it should be said far more loudly. If no, the question is not an AI problem: it is a proposition problem, and it existed before.
A second question, more entertaining and just as serious: which friction have we removed that our customers now rebuild by hand, with paper? That list is worth three satisfaction studies.