How we made our app 80% faster, a data structure story

Birdie builds software for home care agencies, including an app carers use to manage visits, tasks, medications, and observations. Because carers may work where mobile signal is weak, critical data needs to be available offline. Faster performance helps carers spend less time using the app and more time caring for clients.

The team tried common React Native improvements, including reducing re-renders, using VirtualizedList, freezing navigation screens, and upgrading dependencies. The app seemed a little better, but without monitoring they could not measure the change. Manual testing showed that larger agencies tended to have slower performance.

The team suspected that large task data and Redux Persist stringification were slowing the app, so they explored WatermelonDB and later made task downloads optional. These changes took months and did not resolve the slowdown. Datadog RUM gave the team a way to measure real user performance and test its assumptions.

The actual bottleneck was a flat array of user permissions that had grown from about 15 to more than 10,000 per user. The team changed it into an object keyed by agency or client, reducing client permission lookup from scanning up to 10,000 entries to finding the relevant key and checking a shorter list. Switching agencies took about four seconds after the change, compared with 40 seconds before.

Standard mobile vitals did not reveal the difference, so the team added timing for Redux selectors, reducers, and sagas to Datadog. After the permission change, average durations fell by up to 92% for selectors, 90% for reducers, and 64% for sagas. Low-end devices saw the biggest gains, though high-end devices improved too.