PMs WHO BUILD

Y / Build · Week 01 · L3 · 90 min

How to analyze customer feedback with AI: build a triage tool in 90 minutes

Sorting feedback by hand is slow, and pasting it into a chat window gets you confident themes you can't check. This brief has you build a small triage tool instead: it groups feedback by theme, counts each one, flags what's urgent, and cites the rows behind every claim so you can decide whether to trust it. Any tool works. It takes about 90 minutes.

Shannon Owens ·

The brief: Turn a pile of customer feedback into a triage tool. L3 on the Builder PM Ladder, about 90 minutes.

1. Write the bet

PMs lose hours each week sorting raw feedback by hand. A small tool that groups feedback by theme and flags urgent items would cut that to minutes, and you'd trust it enough to use it.

2. Ask for an outcome, not an app

Paste in 50 to 200 rows of real feedback (support tickets, survey answers, app reviews). Get back themes with counts, the three most urgent items with the reason they're urgent, and every item tagged so you can check the work.

3. Catch the failure

Watch for invented themes with no rows behind them, and for one angry customer being counted as a trend. Ask it to show the row behind every claim.

4. Add the guardrail

Make the tool cite row numbers for every theme and refuse to call something urgent without a quoted line. Strip names and emails before you paste anything in.

5. Watch real use

Run it on this week's feedback and compare it with how you'd have sorted it. Where did it beat you? Where would you not trust it?

Tools

Any: Claude, ChatGPT, Lovable, v0, Cursor, Replit

Share what happened

Share a screenshot (no customer data), what you built it with, and one thing that surprised you.

This was the Week 01 brief. This week's is Turn a rough sketch into a prototype one person can click. See every brief so far.