The differentiator was right. The road to it wasn't.

The problem

Leadership wanted one thing and said it in plain words: more sales. The number holding up that conversation was harsh. Out of every ten people who opened the app, barely one finished a purchase. Conversion sat at 12%, measured in Grafana, and nobody could say where exactly the funnel was leaking. There were hypotheses in every direction: price, delivery, catalogue, the app being heavy.

The context made the question harder to answer. Ten developers, no PM, no product organisation and me as the only designer, wearing every hat at once. With no instrumentation and no discovery routine, the decision about what to build next sprint came out of opinion, and opinion is cheap when the funnel has been leaking for months.

The research

I started with whatever would give me evidence fastest. I ran a heuristic evaluation of the app's 32 screens using Nielsen's ten heuristics, with a severity scale from 0 to 4. That produced 80 findings, which I consolidated into 60 unique problems: 14 of them priority and 3 catastrophic. The map by area is what changed the conversation. The cart and checkout alone held 34 of the 80 findings, and every catastrophic one lived there.

I cross-checked that against what already existed and what was missing. I watched session recordings, read churn and bounce data in Grafana, and put together an extensive competitive benchmark against the established players. I also wrote the company's first UX persona, which simply did not exist. The benchmark confirmed the house differentiator was real: no competitor compared the same product across several markets to find the cheapest one.

No competitor compared the same product across several markets. Even so the customer left before ever seeing that work, because getting there meant learning an app that looked like nothing else they had used.

The recommendation

I brought leadership a plan on four fronts. Fix the 14 priority problems starting with the cart, where the fix had the biggest immediate return. Refactor the whole app on top of a new design system, going straight at consistency, which was the most violated heuristic with 19 findings. Start tracking telemetry and product metrics as a routine. And go into the field to interview customers inside the supermarket. I also recommended we stop accepting sign-ups from people outside the delivery area, which was filling the base with people who could never buy.

The design system was born to solve the biggest problem the evaluation itself had pointed at. It landed at 26 React Native components on top of Gluestack UI, with the tokens shipped as a separate package with a primitive and a semantic layer, serving as the single source of truth for the apps and the admin. Two fronts leadership did not buy: widening the service area, and shifting focus from the large retailers to small and mid-sized regional business owners.

The result

Sign-up was the first flow to be rebuilt, because it is the front door to everything that comes after. Drop-off in that flow fell by 32%, comparing the funnel in Grafana before and after the refactor. In usability testing with the new flow, the success rate on the sign-up task landed at 90%, with people finishing on their own, without help and without backtracking.

On the team side, the design system stopped being a promise and became the only path forward: 100% of the refactored screens were built with components from the package, and consistency, the most violated heuristic in the app, dropped off the top of the list. Telemetry came in alongside it, which was the least visible ask in the plan and the one that changes the following year: the next decision gets to have data behind it instead of a hunch.

What I took away

What I take from this cycle is not a course correction, it is the confirmation of a method. It was the first time I walked the whole path in the right order, from finding to metric: heuristics, benchmark, data, persona, system, refactor, testing. If I started over today I would do it the same way, and that is what I look for in any product I pick up from here on.

The part that did not land was the part touching the core business. The proposal to turn focus toward small and mid-sized regional business owners was well received by the people running day-to-day operations and blocked at the leadership level. I learned a distinction there that I have used ever since: evidence settles a product argument, but changing the business model is a political decision, and it needs a different kind of groundwork, slower and with more people on board.