UniCredit builds applied AI into how it decides which product to put forward
A bank's product set barely moves. Cash credit, mortgages, cards — the same shelf, quarter after quarter. What moves is the reason to want them. A cash credit means one thing in a month when people are pricing kitchen renovations and something else when they are working out school costs. The product is identical. The relevance is not, and it changes weekly.
A quarterly plan cannot keep up with that. UniCredit now makes the decision continuously: applied AI reads what the market is thinking about, matches it to the products already on the shelf, and produces the work — with every asset going through the bank's own approvals.
What changed
25% higher click-through on pages built this way
Higher-quality traffic from paid
Lower bounce rates, longer sessions
Same products, same market, same approvals
Our Approach
Financial marketing usually describes the product: the rate, the terms, how it compares. That is accurate, and it is almost never what someone is thinking about. They are thinking about the car, or the kitchen, or the month with too many bills in it.
So the commercial decision is not which product to advertise. It is which product belongs in front of which moment — and that pairing changes faster than a planning cycle. Applied AI holds the product set against what the market is actually doing, and surfaces the pairing while it still holds.
Approvals govern what runs, not when you decide
Every bank assumes this is closed to them. Everything is approved in advance, so nothing can respond to what is happening now.
That reads the constraint too widely. Approval governs what may be published. It does not require the timing of the decision to be fixed months ahead. Recommend supplied the intelligence, the matching and the creative; UniCredit's own team approved and activated under their own compliance process.
Nothing bypassed anything. The work simply arrived while the need was still live.
What it changed
Click-through rose 25% on the pages built this way, with higher-quality traffic and longer sessions alongside it.
The lift did not come from a better offer or a bigger budget. Same products, same market, same approvals. It came from closing the distance between a need existing and the right product appearing.
Built to get sharper
Each product-to-moment pairing carries its own result. That makes this an unusually clean place for the system to learn — the products hold still while the contexts move, so what worked is attributable to the pairing rather than to a change in the offer.
That is the direction the capability runs in, and the reason this is a system rather than a campaign: the record of which pairings landed is what the next quarter starts from.