Reworked first steps in a therapy app: 25% better retention

Credo Therapies is a company that came out of the University of Oxford. It offers a digital CBT-E therapy programme for binge eating disorder, on mobile and web. I built personalised suggestions on top of a language model. The feature had to be quick and its cost predictable. I also had to fit the suggestions into the few steps a new user sees the first time they open the app.

I had to rework new users' first steps so that more of them kept using the app, and fit the AI suggestions in at the same time.

• I connected GPT-4 to the app through server-side functions. I watched the response time and put a limit on how much text the model could take in and send back.
• I simplified the first steps for new users so they would start using the app more easily. For the related server calls, I defined the data types up front.
• I built the release process: Fastlane, GitHub Actions, TestFlight and Play Console.
• I led the WCAG AA accessibility audit and the fixes that followed.

The server calls the language model, so the access key never reaches the user's device

It is the only reliable way to enforce a cost limit and protect against misuse.

• After I reworked the first steps, 25% more users carried on using the app.
• The app’s build process was 40% faster.

• —If you want an AI feature, I plan up front how we measure what it costs to run, how fast it responds and what people use it for.

• credotherapies.com ↗