Why
Our research and prototyping took longest at the points where AI is strongest: sifting through recordings and interview notes, and turning ideas into something testable. Faster there meant testing more ideas with users in each sprint.
How I got buy-in
- Start small: one project, a few clear tasks, and a comparison with how long they took before.
- Agree the rules first: no identifiable patient data goes into AI tools. Recordings and notes are anonymised first, and only approved tools are used.
- Share the results, what worked and what didn't, with the wider team before going further.
Where AI fits now
| Stage | How AI helps | What stays human |
|---|---|---|
| Research | Sifting anonymised Hotjar sessions and interview notes for patterns; first drafts of journey maps | Deciding what the evidence means, and checking every insight against it |
| Ideas | Exploring layout and copy options quickly | Choosing what's right for patients and clinics |
| Prototyping | Turning designs into working coded prototypes for testing | Running the tests and listening to users |
| Copy | Plain-language drafts and versions to compare | Tone, clinical accuracy and the final wording |
| Accessibility | Automated contrast checks and an assisted first review | Testing with people who rely on it |
| Hand-over | Component documentation and token checks against code | Reviewing with engineers |
What changed
Research synthesis and prototyping got faster, so each sprint could test more with users and get changes to market sooner.
What I'd tell another team
Start with the dull, time-consuming parts of the process, not the creative ones. Set the data rules before anyone starts. And keep people deciding: AI drafts; designers and users decide.
Each case study also ends with how I'd use AI on that project today.