Spending Insights
Defined the zero-setup category system and the roadmap behind it.
Now live for millions of current account customers across Lloyds, Halifax and Bank of Scotland.
Theme: Money management
Role: Product definition
Status: Live
2021–22
The problem
Customers told us they were anxious about money — but their banking app couldn't answer the most basic question they had: where does it all go? A statement is a raw list of transactions; turning it into understanding was work the customer had to do themselves. Meanwhile Monzo, Emma and Snoop had already made automatic spend categorisation table stakes, and one of the UK's largest banks — serving tens of millions of customers across three brands — had nothing.
The technical problem was solvable. The design problem was harder: what categories, how much should the customer have to do, and what ships first versus later? The brief: define what spend categorisation should be for the bank — at launch and everything after.
Fig 02
RECREATED ARTEFACT
Customer outcomes: The two customer outcomes the brief was written against — the framing every scoping decision was then measured against.
My role / the team
I was the product designer on Run Ahead, the bank's future-thinking team, and I owned the product definition: the category system, the launch scope, and the 38-feature roadmap that sequenced everything after it. I designed and validated the pilot screens using my own account data. A separate delivery team took the definition into final production and shipped it across all three brands — the product decisions are mine; the final pixels are theirs.
What the evidence said
Three findings constrained the design. First, customer research pointed to around seven categories as the ceiling of comprehension — enough to be meaningful, few enough to read at a glance. Second, configuration is where money-management features die: every setup question is a reason to abandon, so the feature had to be useful with zero input. Third, generic taxonomies fail a simple test — customers must immediately recognise their own spending in the categories, not a textbook classification of spending types. Together these ruled out both build-your-own categories and a long ML-driven taxonomy, and pointed at a small, fixed, universally legible system.
The judgment calls
Zero setup beat personalisation.Three ways to categorise spending: let customers build their own categories (accurate, but most never finish setup), use ML to personalise (opaque, and every misfiled transaction erodes trust), or fix a small universal set. I chose eight fixed categories that every customer — one current account or five products — would recognise without doing anything. What was traded away: day-one personalisation. What was gained: a feature that was useful in the first ten seconds, for everyone.
"Everything else" is an honest bucket.The tempting move is to force every transaction into a named category so the system looks omniscient. Instead, edge cases went into "Everything else" — trading taxonomic completeness for the integrity of the other seven. If Groceries says Groceries, it's all groceries. The side effect turned out to matter: the bucket quietly surfaces the tail of miscellaneous spending customers hadn't noticed.
Ship seven features; sequence thirty-one.The full vision — budgets, custom categories, income comparison, proactive nudges — was a 38-feature roadmap mapped to two customer outcomes: understand where your money goes (launch) and make better decisions with it (future state). I scoped launch to seven features serving the first outcome only, trading feature parity with the challenger apps for a pilot that could ship, learn from a built-in feedback loop, and earn each next phase. The roadmap made the deferrals a plan rather than an omission.
The design
These are the pilot screens: prototyped against real transaction data, tested, and handed to delivery largely as they stand. The annotations mark the seven features that made the launch cut — and the feedback loop built in to check the categories were right.
Fig 03
RECREATED ARTEFACT
Pilot design: Five of the seven launch features in one view — plus the feedback loop that asked whether the categories worked.
Fig 04
RECREATED ARTEFACT
Pilot design: The remaining two — month-on-month comparison, and every visit to one supermarket rolled into a single figure.
What shipped & what happened
The launch pilot shipped with real customer data and the feature went live across Lloyds, Halifax and Bank of Scotland — in front of millions of customers. The pilot covered the full state model: current and historic months, merchant drill-downs, empty, loading and error states, and the category feedback form.
What I can't show: post-launch engagement or satisfaction data — I moved on before it was gathered, and it isn't mine to share. What I'd measure: the feedback form's yes-rate on category recognition, repeat monthly visits, and drill-down depth as a proxy for genuine curiosity about spending. Qualitatively, the category system survived the journey from recommendation to production largely intact — the strongest signal the definition work held up.
Fig 05
RECREATED ARTEFACT
Product roadmap: 38 features across four delivery phases, each mapped to a customer outcome — the plan the launch scope was cut from.
What I'd do differently
Stay closer to delivery. The gap between recommendation and shipping was over a year, and handoff is where product intent leaks — an advisory role through build would have protected more of it. I'd also have pushed harder for commitment to Day 2 and 3, where a read-only view becomes a genuine money-management tool.
Some artefacts are shown at reduced fidelity, others redrawn. Figures are illustrative.