Verified against Claude · 2026-08-03
Find out why customers actually left, not just which canned reason they clicked
Synthesizes churned-customer exit notes by categorizing free-text reasons and separating never-activated churn from used-then-quit churn — the two need different fixes, and lumping them together produces mushy insight.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are synthesizing churn exit interviews and cancellation notes. Your job is to categorize what customers actually said in their own words, not to default to whichever canned multiple-choice reason they happened to click on their way out the door. CONTEXT Number of churned customers in this batch: 14 Raw exit notes/interview transcripts, labeled by customer if possible: Customer 3 (active 11 weeks): "It worked fine, honestly, I just realized I was the only one on my team still using it after the new hire started"... Customer 7 (active 2 days): "Never got past connecting our POS, gave up"... Canned cancellation-flow reasons selected, if collected separately from free text: Customer 3 selected "Too expensive." Customer 7 selected "Missing a feature I needed." Time each customer was active before churning, if known: Customer 3: 11 weeks. Customer 7: 2 days. ACTIVATION SPLIT First split the batch into two groups using Customer 3: 11 weeks. Customer 7: 2 days. and the notes: customers who never meaningfully activated (never reached real usage of the core value) versus customers who used the product for a real stretch and then quit. These are different problems — one is an onboarding failure, the other is a retention failure — and must be analyzed and reported separately, never merged into one combined churn theme. FREE-TEXT OVER CANNED For each group, extract the real reason from the free-text notes in Customer 3 (active 11 weeks): "It worked fine, honestly, I just realized I was the only one on my team still using it after the new hire started"... Customer 7 (active 2 days): "Never got past connecting our POS, gave up"..., not from Customer 3 selected "Too expensive." Customer 7 selected "Missing a feature I needed." alone — a canned reason list anchors people toward whichever option is easiest to click, and the free text underneath frequently tells a different, more specific story. Where the canned reason and the free text disagree, report the free text as the primary signal and note the mismatch explicitly. PATTERN RULE Within each group, promote a reason to a real theme only if it appears in at least 2 customers' notes. A single vivid, detailed complaint is not a pattern by itself — label it "Isolated case" and keep it visible without treating it as representative of the batch. OUTPUT FORMAT Two headed groups (Never Activated, Used-Then-Quit), each with themes (count + representative quotes) and isolated cases, plus a short note on any canned-reason mismatches found.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Many churn exit surveys default to a multiple-choice reason list, and that list has a documented and specific bias: it anchors respondents toward whichever option is closest to their actual reason without being it exactly, because clicking a close-enough checkbox takes ten seconds and explaining the real, more complicated story takes longer than a departing customer is willing to spend — which is exactly why {{canned_reasons_selected}} and the free text underneath it frequently disagree, and why the prompt treats the free text as the primary signal rather than the canned click. Separating never-activated churn from used-then-quit churn matters because these are structurally different problems needing different fixes: a customer who never got past connecting their POS has an onboarding failure, and a customer who used the product for 11 weeks and then quit because their team stopped needing it has a retention or product-fit failure — lumping both into one 'churn reasons' theme produces a mushy, average-of-two-different-problems insight that doesn't point clearly at fixing either one. The n≥2 pattern rule applied to negative signal specifically matters because churn stories are unusually easy to over-react to — a single vivid, detailed complaint about a missing feature reads as urgent and specific in a way that makes it tempting to treat as representative, but without the same discipline applied to positive-signal synthesis (require at least 2 independent instances before calling it a theme), a team can end up rebuilding a roadmap around one departing customer's particularly well-articulated complaint.
Verified against
Claude Sonnet 5 · 2026-08-03
ChatGPT GPT-5.1 · 2026-07-30
Changelog
- 2026-08-03 — Initial publish, verified against Claude Sonnet 5 and ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — custom software, built and maintained for you — that's Scult's day job.
EXPLORE CUSTOM SOFTWARE
