The Churn Autopsy Prompt: Turn Exit Interviews Into a Product Roadmap

Every cancelled account is a free consulting session you already paid for. Most CS teams sit through the exit interview, file the notes somewhere, and learn nothing from it. Six months later, the next account churns for the exact same reason, and nobody connects the dots — because the notes were unstructured, the themes were never counted, and product never saw a version of the data they could act on.

The churn autopsy prompt fixes all three problems at once. Feed it your exit interview notes — a month, a quarter, even a year of messy notes from different reps — and it returns product gaps ranked by the revenue each one cost you. Not sentiment. A product roadmap input with dollar signs attached.

Here's how to run it, plus the exact prompt to copy.

Why exit interviews usually teach you nothing

The failure isn't the conversation. Most CSMs ask decent exit questions. The failure is what happens after.

First, the notes live in a CRM text field nobody reads. "Left for competitor, price too high, champion departed Q2." That's data the way a junk drawer is storage — technically contained, practically useless.

Second, every rep writes notes differently. One person's "integration issues" is another person's "API too hard to implement." You can't aggregate what isn't consistently described.

Third, the reasons customers state are rarely the reasons they churn. "Price too high" from a customer who used 10% of the seats usually means "we never got value." The stated reason is the headline; the real reason is two questions deeper, and the note-taker didn't always dig.

The prompt below is designed around all three failures: it normalizes inconsistent notes, looks past stated reasons to root causes, and outputs product gaps ranked by revenue impact.

Step 1: Gather the raw material (30 minutes)

Pull every exit interview note, cancellation survey response, and churn reason field from the last 90 days. Don't clean them. Paste them into a single document with one account per block, including the account's ARR, the stated reason for leaving, and the raw notes. Ten accounts is enough to get signal. Thirty is better.

If your notes are thin — a CRM dropdown and nothing else — the prompt still works, but the output will tell you your interview process needs work. That's useful information too. Treat thin output as the trigger to fix your exit interview script (covered at the end).

Step 2: Run the churn autopsy prompt

This is the prompt. Paste your raw notes where indicated and run it in any frontier model.

The Churn Autopsy Prompt

ROLE: You are a customer success analyst performing a churn autopsy. Your job is to convert messy exit interview notes into a prioritized product roadmap input.

INPUT: Below are exit interview notes from [N] cancelled accounts. Each block includes the account name, ARR at churn, stated cancellation reason, and raw notes from the CSM.

[PASTE NOTES HERE]

TASK: Analyze all accounts and produce the following:

1. Root-cause table. For each distinct reason customers churned, list: (a) the root cause in plain language, (b) the stated reason it hid behind, (c) the number of accounts affected, (d) total ARR lost to this cause, and (e) one representative quote from the notes. Separate stated reasons (what the customer said) from root causes (what actually drove the decision, inferred from evidence across accounts).

2. Product gap ranking. Translate the root causes into specific product gaps or missing capabilities. Rank them by a simple score: (accounts affected × 2) + (ARR lost ÷ $10,000). For each gap, write: the gap in one sentence, the score, how many accounts it touched, and what a "good enough" fix would look like in under 60 days.

3. Non-product churn drivers. List every root cause that is NOT a product gap — onboarding failures, pricing/packaging issues, support experience, champion turnover, acquisition mismatch. For each: the driver, accounts affected, ARR lost, and the one process fix that would prevent the most recurrences.

4. The uncomfortable finding. Name the single churn driver your CS team is most likely to be avoiding or under-reporting, based on patterns in the notes (thin documentation, vague reasons, rep blame-shifting language). Be direct.

5. Top 3 asks for product. Your final output: three specific, evidence-backed product asks, each with the revenue it would have saved last quarter and the quote that makes the case. Write them the way a PM would write a ticket — problem, evidence, proposed scope.

RULES: Do not invent reasons the notes don't support — flag thin evidence. Do not double-count one account under multiple root causes unless the notes clearly show two independent causes. Keep the analysis under 800 words.

Step 3: Read the output like a skeptic

The prompt's most valuable line isn't any of the rankings — it's the rules section at the end. AI will invent tidy causal stories from thin evidence if you let it. The "do not invent reasons" rule forces it to flag weak evidence, which is your signal to go re-interview those accounts or accept that your data has holes.

Watch for three things. First, the stated-vs-root gap. If "price" is the stated reason but the root cause is "never reached first value," your problem isn't pricing — it's onboarding, and a discount would have saved nobody.

Second, the uncomfortable finding. In every team I've seen run this, section 4 lands. It's usually something like: reps are writing "customer went dark" instead of admitting they stopped calling, or churn is being attributed to "budget cuts" in accounts where usage flatlined months before the budget conversation. The prompt finds the pattern because it reads across notes that no single human ever reads together.

Third, check whether section 5's top asks surprise you. If they do, that's the point — you now have evidence-backed asks instead of the loudest CSM's opinion. If they don't, that's also the point — you now have the revenue math to get product to act on what you already suspected.

Step 4: Hand product the three asks (this week, not next quarter)

The output of section 5 is a product-ready artifact. Send it to your product counterpart with this framing: "Here's what churn cost us last quarter, broken down by the product gaps behind it, with customer quotes. These are the three fixes that would have saved the most revenue."

Product teams ignore feedback that arrives as vibes. They act on feedback that arrives as ranked, revenue-weighted, quoted evidence. The format matters as much as the content — PMs think in tickets, so give them tickets.

Then do the part everyone skips: run it monthly. A quarterly autopsy is a post-mortem; a monthly one is a feedback loop. Save each month's output and watch the gap rankings shift — the gap that falls off the list is your proof that a shipped fix worked, which is the best slide you'll ever put in a QBR.

Fix the input while you're at it

The prompt gets better as your notes get better. If the output was thin, fix the exit interview itself. Three questions, asked every time, in this order:

1. "What was the moment you decided to leave?" Not why — when. "When our champion left in March and nobody reached out" is a process failure, not a product failure.

2. "What would have had to be true for you to stay?" This forces specificity. "Lower price" is a dodge; "if the sync worked without our IT team" is a product ask.

3. "What's the one thing we did well?" Ending on strength keeps the door open for a win-back and tells you which parts of your experience are worth protecting while you fix the rest.

Write the answers down verbatim. Future you — and this prompt — will thank present you.

The bottom line

Churned customers already did the research for you. They found every weakness in your product, onboarding, and support, and told you about it on the way out. The churn autopsy prompt is just the discipline of actually listening — systematically, monthly, with revenue attached to every finding.

Pull last quarter's exit notes. Run the prompt. Send product the three asks. Do it again next month, and watch the list shrink.

Tomorrow: "No More 'What Time Works for You?': Async-First Scheduling for Global CS Teams" — Loom-first check-ins, timezone-proof handoffs, and when async fails.