When to use it
When customers or subscribers are leaving, you don't know why, and you need to hold on to them. The role: retention product analyst. The result: hypotheses about causes across the lifecycle, segments showing who churns most, early risk signals, and a retention plan with quick wins and a metric.
The prompt (copy and paste)
You are a retention product analyst. Help me break down churn. Reason step by step, don't give one shallow answer.
PRODUCT: "<FILL IN: what the product is, the model — subscription/one-off purchases>". CHURN: <how many leave, over what period, if I know>.
WHAT I KNOW: "<data, feedback from people who left, where they drop off, exit surveys>".
Do this:
1. Split the customer journey into stages (onboarding → first value → habit → renewal) and for each, name why people might leave there.
2. Hypotheses for churn causes, most likely first — each tagged with the data that would test it.
3. Risk segments: who churns more often (by source, plan, behaviour) — if the inputs show it.
4. Leading indicators — what signs show a customer is about to leave, BEFORE they do.
5. A retention plan: 2-3 quick wins (shippable in a week) plus 2-3 systemic fixes, each labelled with the cause it addresses and the metric to watch.
If there isn't enough data, say what to collect (exit survey, cohorts, event analytics) rather than inventing causes.
Filled-in example
Product: a SaaS booking subscription for salons. Churn: 8% a month. Known: many leave in the first two weeks, exit surveys are rare.
What the AI should return: stages and risks (onboarding — "never set up their schedule → never saw the value"); hypotheses — "onboarding is too complex", "they never migrated their client list", "no integration with what they already use", each with a test; segments — solo practitioners churn more than salons; early signals — "hasn't added a single booking in 3 days", "hasn't logged in by day 5"; the plan — quick wins (an onboarding checklist, a day-3 email for inactive users, help migrating their client list), systemic fixes (integrations, tracking "first value"), metric — week-2 retention and the share of users who reach their first booking.
Variations
- Exit survey. "Write a short survey for people who are leaving — 3-4 questions that get at the cause without annoying them."
- Cohorts. "How do I break retention down by cohort, and what do I look for in the chart to find the drop-off point?"
- Win-back. "A win-back sequence for churned customers: who to contact, with what offer, and when it's appropriate."
Pro tips
- Fight for early signals: you can only retain someone BEFORE they leave. "Didn't reach first value within N days" is the most common and most fixable signal — find yours.
- Onboarding churn and long-tenure churn are different diseases with different cures. Make the AI split by stage instead of treating "average" churn.
- The AI gives you hypotheses and a plan, but causes are confirmed by data and by talking to the people who left. Don't paste a customer database with personal data into a public AI — work from aggregates and anonymised exports.