What it does
Users build a shelf of the skincare they own. The app prices every product across every stockist and tells them what they could be saving: 38 products, £1,186 RRP, £141.90 available saving, on one screen. Product pages show live prices from each retailer with a Buy Now that goes to the cheapest. A barcode scanner adds a bottle you are holding in a shop.
Then there is a whole social layer: share a shelf by link, follow friends, send and accept requests, match contacts, browse what other people actually use.
What I did
Installs were flat the entire time. The audience still grew 4.2× in monthly actives and 20× in monthly engagement, because the work went into the product rather than the traffic.
- Stale-while-revalidate caching with optimistic mutations. The app renders immediately and reconciles behind you. Weekly actives went up 13× in 90 days.
- Three platform migrations, concurrently. REST v1 to v2, JWT to Laravel Sanctum, search to Typesense. Zero downtime, no feature freeze.
- SOLID refactor. 25 BLoCs moved to constructor-injected dependency injection, full test suite green throughout.
- Attribution and observability. AppsFlyer OneLink deep links, App Tracking Transparency, PostHog, Crashlytics. This is what produced the +192% in attributable monthly user spend.
- First mobile CI pipeline for the team.
- Product variant inference, so one item stops appearing as five.
- CMS-configurable advertising and spotlight blocks, so the client ships campaigns without a release.
- A long tail of fixes: duplicate friend-request notifications, crash reduction, a search endpoint returning complete results again.
Why it matters
Growth without new installs is the hardest kind. It means the product got better for the people already there. That is the work I like most.
