Cancel-flow text is the most under-mined dataset in subscription products. Every response carries two things: the reason a user gives (usually polite) and the reason behind it (the real driver). These 15 prompts cluster cancel reasons by root cause, split reversible churn from inevitable, find the lifecycle stage where churn concentrates, and turn the output into save-flow copy and product-fix priorities. This set is specifically about parsing voluntary-exit text — not the cohort curves and metrics of general churn analysis.
TL;DR
- Run the prompts on a model with a large context window so you can paste hundreds of verbatims or whole interview transcripts in one shot: Claude Opus 4.7 or Sonnet 4.6 (1M tokens), or Gemini 3.1 Pro (1M tokens) all handle the volume as of June 2026.
- Cluster by the free-text explanation, never the dropdown label the user clicked.
- Tag every reason reversible / inevitable / misaligned before you size any intervention.
- Don’t ignore involuntary churn (failed payments). It is roughly a quarter of total churn and the cheapest to recover — dunning routinely reclaims 50-80% of failed payments.
- Treat exit-survey data as a lagging signal. Most churn started weeks earlier in behavior (billing-page visits, feature abandonment, logins with no outcome), so pair these prompts with engagement work.
Who this is for
Subscription PMs and lifecycle marketers, CX leaders running save-flows, growth analysts doing cohort-churn studies, and founders who want to know what their churning users actually thought before they left.
When not to use these prompts
- Under ~50 free-text responses a month — read each by hand; clustering needs volume to be meaningful.
- Dropdown-only data — a single selected reason (“Too expensive”) tells you almost nothing about the driver. You need free-text or interview transcripts.
- Silent churn (users who never cancel, just stop logging in) — that is an engagement problem, not a churn-reason problem. Use retention-experiment prompts instead.
Churn benchmarks to calibrate against (2026)
Use these as a sanity check on your own clusters. If your involuntary share is far below ~25%, you may be misclassifying failed payments as voluntary cancels.
| Metric | Typical B2B SaaS, 2026 | Source |
|---|---|---|
| Monthly total churn | ~3.5% | culta.ai benchmark |
| Voluntary share of churn | ~74% | shno.co |
| Involuntary share (failed payments) | ~26% (20-40% of losses) | shno.co |
| Failures caused by expired cards | ~42% | shno.co |
| Failed-payment recovery with good dunning | 50-80% (top performers 80-90%) | Churnkey / Baremetrics |
These figures are directional industry averages, not guarantees — your numbers will move with segment, price point, and billing setup. (See the Sources below.)
Which model to run these on
These are qualitative-synthesis prompts: the bottleneck is context length and reasoning over messy text, not code or speed.
- Long transcripts or thousands of verbatims — paste them whole into Claude Opus 4.7, Claude Sonnet 4.6, or Gemini 3.1 Pro (each ships a 1M-token context as of June 2026). Note that ChatGPT Plus exposes only ~320 pages of in-app context; the full 1M window is on the $200 Pro tier.
- A quick weekly cluster of a few hundred responses — Sonnet 4.6 or Gemini 3.1 Pro is faster and cheaper. Per the canonical API rates (USD per 1M tokens): Sonnet 4.6 is 3/15, Gemini 3.1 Pro 2/12, Opus 4.7 5/25.
- Reserve Opus 4.7 for the harder reasoning passes — reason-behind-the-reason inference (template 6) and the predictive-signal extraction (template 15), where the deeper model earns its premium.
How a churn-reason prompt should be built
Skip the generic six-part formula. For cancel-flow analysis the four parts that actually change the output are:
- Cluster instruction — say “cluster by the underlying driver, ignore the dropdown label” explicitly, or the model mirrors the multiple-choice options back to you.
- Classification axis — force a reversible / inevitable / misaligned tag on every cluster so the output is decision-ready, not just a list.
- Evidence requirement — demand 2-3 representative verbatim quotes per cluster so a skeptical exec can audit the grouping.
- Counter-move — ask for one concrete next action per cluster (product fix, pricing test, dunning change, or “do not chase”).
15 copy-ready prompt templates
Replace every [bracketed] placeholder with your data. The templates are written for any of the long-context models above; swap the role line if you want a more domain-specific voice.
1. Cancel-flow free-text clustering
The default. Cluster cancel reasons by ROOT driver, not surface answer.
You are a retention analyst. Below are [N] free-text cancel-flow responses for [product]. Cluster into 6-8 root reasons (not surface labels). For each: count, % of churners, 3 representative verbatim quotes, classification (reversible / inevitable / unclear), and one suggested counter-move.
Responses: [paste]Optimization: If clusters mirror your dropdown labels, add: “Ignore the multiple-choice labels users selected. Cluster only by the free-text explanation.”
2. Reversible vs inevitable split
From this churn analysis, split reasons into 3 buckets: (1) reversible — we could prevent with product / pricing / support work, (2) inevitable — life event, role change, project ended, (3) misaligned — we should not have sold them in the first place. For each bucket: count, % of churn, recommended action.
Clustering: [paste]3. Stage-of-lifecycle churn map
Below are churned users with tenure (days since signup). Map churn reasons against lifecycle stage: early (0-14d), trial (14-30d), early-paid (30-90d), mature (90d+). Output a stage-by-reason matrix. Highlight which stage produces the most reversible churn.
Data: [paste]4. Exit-interview transcript synthesis
Below are 10 exit-interview transcripts (15-30 minutes each). Synthesize: top 5 themes, 3 surprising findings, 2 things multiple users said that contradict our current beliefs. For each finding, cite which transcript supports it and a recommended action.
Transcripts: [paste]5. Downgrade-trail diagnostic
Some users downgrade before churning. Analyze the downgrade trail: average days from downgrade to churn, common reasons given at downgrade, and whether a save attempt at downgrade prevents the eventual cancel. Output a trail timeline plus the single best intervention point.
Data: [paste]6. Reason-behind-the-reason extraction
Users often give polite cancel reasons that hide the real one ("price" frequently means "not getting enough value"). For each of these 5 verbatim responses, infer the likely underlying reason, mark your confidence, and state the one follow-up question that would confirm it.
Responses: [paste]7. Save-flow message generator
For each of these 6 cancel reasons, write 2 save-flow messages: (a) a soft acknowledge plus offer (extend trial, pause, talk to founder), (b) a no-pressure exit (thanks plus feedback ask). Voice: respectful, never desperate. Under 60 words each.
Reasons: [paste]8. Persona-skewed churn detection
Compare churn reasons across user personas: free-to-paid, paid-power, paid-occasional, enterprise. Highlight reasons that skew sharply by persona — those become persona-specific interventions rather than blanket fixes.
Data: [paste]9. Competitor-loss reason extractor
From these cancel-flow responses, extract every mention of a competitor ("switching to X", "X has better Y"). For each: which competitor, what feature they cited, frequency. End with the top 3 competitor-driven losses and what we would need to ship to close that gap.
Responses: [paste]10. Pricing-driven churn diagnosis
For users citing pricing as the cancel reason, diagnose the sub-cause: absolute price (too expensive), perceived value (not enough for the price), pricing structure (wrong tier), or budget cycle (price unchanged, budget shrank). Recommend a different counter-move per sub-cause.
Responses: [paste]11. Quarterly churn-reason delta
Compare last quarter's churn reasons to this quarter's. Output: reasons that grew, reasons that shrank, new reasons, vanished reasons. Hypothesize why each shifted. End with the 3 themes worth investigating next quarter.
Q-1: [paste]
Q0: [paste]12. Voluntary vs involuntary churn separator
Below is our churn dataset including both voluntary cancels and involuntary churn (failed payments, expired cards). Separate them. For involuntary churn, analyze recovery rate by retry strategy. For voluntary, cluster the reasons. Output two parallel reports.
Data: [paste]13. Save-rate by intervention test
We ran 4 save-flow interventions over the last 90 days: pause, discount, talk-to-founder, content-only. Measure save rate per intervention, durability of saves (still active at 60 days), and cost per save. Recommend which interventions to keep, modify, or kill.
Data: [paste]14. Refund-request pattern audit
Below are refund-request reasons over 90 days. Cluster by root cause. Flag any pattern where refund requests cluster within 7 days of signup — that signals an onboarding or messaging problem upstream of billing.
Requests: [paste]15. Predictive churn-signal extractor
From this dataset of churned vs retained users, identify 5 leading indicators of churn in the 14 days before cancellation. For each signal: how to detect it programmatically, what intervention to trigger, expected save rate. Mark which signals are causal vs merely correlative.
Data: [paste]Common mistakes
- Clustering on the dropdown label users selected instead of the free-text they wrote.
- Treating “too expensive” as a price problem when it usually means “not enough value” — fix template 10 will separate the two.
- Ignoring involuntary churn. It is ~26% of total churn in 2026 benchmarks and ~42% of failures are just expired cards; dunning recovers most of it.
- Building save-flow offers without testing whether they retain durably (still active at 60+ days). A 90-day-only save can be a worse outcome than letting the user go.
- Acting on a single emotional exit interview instead of the cluster.
- Missing the lifecycle stage where churn concentrates — early-trial and mature churn need different interventions.
- Confusing reversible and inevitable churn, then burning the team chasing the latter.
How to push results further
- Require free-text in the cancel flow — dropdown-only data is nearly useless for this analysis. Keep the survey to 3-5 questions so users still answer.
- Cluster by root cause, not surface label, every time.
- Tag each reason reversible / inevitable / misaligned before sizing the intervention.
- Track save durability at 60-90 days, not save rate at the cancel screen.
- Pair quantitative cancel-flow data with 5-10 exit interviews per quarter — numbers tell you what, interviews tell you why.
- Write down a “things we will not save” policy so the team stops chasing inevitable churn.
- Watch the quarterly delta in reasons (template 11). Drift signals a product, market, or pricing shift before the metrics catch up.
FAQ
- How does this differ from general churn analysis?: General churn analysis covers cohorts, retention curves, and metrics like NRR. This set is specifically about parsing voluntary-exit text — cancel-flow free-text, exit interviews, downgrade trails — to extract actionable reasons.
- How much cancel-flow text do I need?: Minimum ~50 free-text responses for clustering to surface real patterns. Below that, read each by hand. Volume is also why you want a 1M-token model — you can paste the whole month at once.
- Which model should I use?: Any of Claude Opus 4.7, Claude Sonnet 4.6, or Gemini 3.1 Pro (each 1M-token context as of June 2026). Use Sonnet 4.6 or Gemini 3.1 Pro for routine weekly clusters, and Opus 4.7 for the harder inference (templates 6 and 15).
- Should I offer a discount in the save-flow?: Only when the cancel reason is specifically absolute price. Discounting a value problem trains users to expect discounts and erodes revenue. Use template 10 to confirm the sub-cause first.
- How do I tell reversible from inevitable churn?: Use template 2. Reversible: product gap, pricing, onboarding, support. Inevitable: life event, role change, project ended. Misaligned: a customer you should never have sold.
- What about silent churn (no cancel, just stops using)?: Different problem — that is engagement and retention work. Use the user retention experiment prompts for it.
Sources
- SaaS Churn Rate Benchmarks 2026 — culta.ai
- Customer Churn Statistics 2026 (voluntary vs involuntary) — shno.co
- Recover Failed Payments — Baremetrics