What each version fixed, what it did not, and what comes next
The first version of PawHealthAI asked six steps of questions before showing anything. I have iterated onboarding across multiple versions since. This is what changed each time, why, and what is still unsolved.
Sole designer and technical owner. Initial launch in 2025; revised onboarding in April 2026. Completed-page saving added in October 2026, live on Android and coming to iOS.
The first release asked owners to complete setup before they could see the app. Pet details make Evie's educational guidance more relevant, but collecting too much upfront creates work before value. The challenge was to change when we ask without losing useful pet context.
Why were owners leaving before they had started?
The first release asked for six steps of pet details before showing the app. Owners were asked to do all the work before they had seen any value.
I cut the upfront form to two steps, so owners reach the app quickly and can see what it does. The other four steps moved inside the app, in front of Evie. I kept them because Evie needs to know the pet to give proper advice, so the assistant stays locked until they are done.
The April 2026 revision moved app entry earlier. The four remaining pages still requested the context needed for Evie. This changed the timing of setup; no validated completion or retention improvement has been established. Historical July records had instrumentation concerns and cannot serve as a comparative funnel.
What happens when someone stops halfway?
The questions stayed the same. Each completed page is saved, and the owner resumes at the first unfinished page. This does not imply that every field on an unfinished page is saved. These diagrams illustrate the implementation; an actual resume recording remains to be added.
Six steps before the app.
Two steps before app access; four further pages inside.
Each completed page saved; resume at the first unfinished page.
Completed-page saving is now live on Android, with iOS to follow. The revised sequence moves app entry earlier, and the latest implementation protects completed progress. I do not have reliable comparative results for either change.
Document-assisted setup is a future product question: could an owner upload a vet record and confirm extracted details instead of entering each one? Reliable event reporting and a forward-looking cohort are needed before comparing completion between versions.
Changing setup timing and protecting completed progress, with explicit evidence limits.
Keeping the detail the product needs while removing the pressure to finish in one go.
Figma flows and React Native implementation, by one person.
Saying what is observed and what is not yet measured.
Co-Founder and CTO, PawHealthAI. Sole designer and engineer of the app and website.
I cover the complete design and implementation story in Designing and building PawHealthAI.