The feature was ready. The users weren't in the room yet.

After a discussion with my product leader about improving user retention between sessions, push notifications became the feature to design. The question wasn't whether to build them - it was how to make them feel personal rather than generic.

I mapped out how the experience should differ across user types:

• How the experience should differ across user goals (weight loss, gain, maintenance)
• How the experience should differ across health conditions and medications
• When a notification should fire vs. stay silent
• What the copy should say for each scenario
• How timing and frequency should work to avoid fatigue

The result was a detailed feature design covering these dimensions across four notification layers.

The feature was ready on paper. But notifications are inherently personal - they depend on who the user is, what they ate that day, what medication they're on. I could see it clearly in my head. I wasn't sure the team could.

I created three synthetic personas grounded in our actual user base:

• Priya, Bangalore, on Metformin, weight loss goal
• James, Vancouver, on Atorvastatin, weight gain goal
• Sarah, Toronto, no medication, maintenance

Using Claude Code to build a browser-based simulator, I could change what Priya ate for lunch and see in real time what notification she'd receive, why she'd receive it, or why she wouldn't. The feature design stopped being a document and started being an experience.

* Representational demo — product name and theme changed for confidentiality.

Talking to real users is always the goal. Using AI to build this upfront meant that when those conversations happened, we already knew what to look for.

• Several scenarios where the experience differed unexpectedly across user types
• The weekly summary as the highest-value piece to prioritise first
• Edge cases that weren't visible reading the feature design flat on a page

The team could see the feature from the user's side. Questions that would have taken multiple back-and-forths resolved in one session.

This also became the starting point for further brainstorming - the weekly notification layer that emerged from this session became the next feature we went on to conceptualise and build out in detail.