01Context
The cards area of the app had grown by accretion over four years. Every new capability — freezing a card, changing a limit, ordering a replacement, toggling contactless — had been added wherever there was room, which usually meant another item at the bottom of another list. The screen worked, in the sense that everything was reachable.
Nobody owned it end to end. Two squads shipped into the same surface on different cadences, and the design decisions that mattered were being made implicitly, in backlog grooming, by people optimising for their own release.
02Problem
The call centre data was the thing that reframed it. Roughly a fifth of card-related calls were for actions the app already supported — people were phoning in to do something they could have done themselves in fifteen seconds, had they known where it lived.
We had been treating this as an information architecture exercise. It was closer to a discoverability and trust problem: even people who found the controls hesitated, because nothing on the screen told them what would happen after they tapped.
People weren't lost. They were unsure — and an unsure person phones the bank.
03Approach
Three moves, and the order mattered more than any one of them:
- Instrument before redesigning. We spent the first six weeks adding events to the existing screens so we could argue from data rather than taste.
- Reframe around intents, not features. The eight most common reasons people opened the card screen became the organising structure, replacing the feature inventory that had accumulated.
- Make consequences visible. Every destructive or semi-destructive action states its effect and its reversibility inline, before confirmation.
04Decisions
The tradeoffs worth defending. Each one cost us something real.
One shared surface over two squad-owned screens
A single card detail screen with one design owner and a written contract for what either squad may add to it.
Letting each squad keep its own section, which was the politically cheaper option and required no reorganisation.
The fragmentation was a reporting-line artefact showing up in the interface. No amount of layout work survives two teams optimising separately. Cost: roughly six weeks of negotiation before a single screen was drawn, and one capability slipped a quarter.
Eight named intents over a personalised feed
A fixed, ordered set of eight intents derived from research and call centre logs, identical for every customer.
Ranking the actions per user based on behaviour, which the data team had already prototyped and wanted to ship.
In a regulated product, a screen that reorders itself is a screen that support cannot talk a customer through on the phone. Predictability was worth more than relevance here. Cost: the eighth intent is genuinely rare and takes space from the first seven.
05Outcome
Self-service card actions rose 64% in the two quarters after rollout, and card-related call volume fell 28%. The number that mattered internally was different: eleven legacy screens were retired, which is what finally made the surface maintainable by one team.
The intent taxonomy outlived the redesign. It's now the structure the roadmap is written against, which I'd argue is the more durable output.
06Reflection
I under-invested in the instrumentation argument early and paid for it. For the first two months I was defending a redesign on craft grounds against people who quite reasonably wanted evidence. The six weeks of instrumentation should have come before the first concept, not alongside it.
I'd also push harder on the ownership question sooner. It was obvious by week three and I spent another two months designing around it rather than naming it.