# Churn Desk > A subscription churn workbench for founders and growth, product or billing leads: paste your churn picture (customers and cancellations, MRR, how many cancels were failed payments, failed payments and recoveries, cancel-flow sessions and saves, cancel reasons with counts, the cancel flow you have today, how failed payments are handled) and your plans and prices, and find where subscribers are leaking - voluntary or involuntary, which cancel-flow stage or dunning step is missing or wrong - then build the reason-matched cancel flow or the failed-payment recovery sequence. The free scan runs in the browser and uploads nothing. The paid audit orders the fixes; the paid flow and recovery lanes design them. App: https://churn-desk.skillsafe.ai/ API tutorial: https://churn-desk.skillsafe.ai/api.html ## Input - The churn picture: plain text, Markdown, CSV, JSON or saved HTML, pasted or dropped. Customers at the start of the period, cancellations, MRR or revenue per customer, how many cancels were failed payments (a count or a percentage), failed payments and how many were recovered, cancel-flow sessions and saves, offers shown and accepted, pauses and reactivations, cancel reasons with counts or percentages under a heading such as "Cancel reasons", the cancel flow as it is today, how failed payments are handled today, the billing provider. - Plans, prices and limits (optional, needed for dollar savings): each plan and its monthly or annual price, whether pause or a lower plan exists, the tone, the most you will discount. - Optional: the business type - B2C / self-serve, B2B / team plans, or freemium (otherwise read from the notes). ## The free scan - The numbers, each with its arithmetic: monthly churn rate (churned / customers at the start, or as stated) against the skill's target (under 5% B2C, under 2% B2B); the involuntary share (failed-payment cancels / all cancels; typically 30-50%); payment recovery (recovered / failed: poor under 30%, average 40-50%, good 60%+); cancel-flow save rate (saved / sessions; 25-35%); offer acceptance (15-25%); pause reactivation (60-80%); revenue per customer (MRR / customers, or the only plan's price); MRR lost to churn and to unrecovered payments. - Cancel reasons mapped to the skill's eight exit-survey categories - too expensive, not using it enough, missing a feature, switching to a competitor, technical issues, temporary or seasonal, business closed, other - with shares and the offer each calls for (price: 20-30% off for 2-3 months, else a downgrade; usage: a 1-3 month pause, else onboarding help; feature: a roadmap preview, else a workaround; competitor: a comparison plus a discount, else a feedback session; technical: escalate to support, else a credit and priority fix; temporary: pause, else a temporary downgrade; closed: no offer). - The five cancel-flow stages (trigger, exit survey, save offer, confirmation, post-cancel) and six dunning steps (pre-dunning, smart retries, decline handling, dunning emails, grace period, card updater), each found, said to be missing, unclear or not found, with the lines that carry them. - What a 20%, 25% or 30% discount saves per month and over 2 or 3 months on each monthly plan, the gap between plans, and annual against monthly. - Flags CH-01 to CH-23: churn above target, cancel by call / email / chat only, guilt-trip copy, discounts of 50% or more, above 30% or longer than 3 months, pauses longer than 3 months, one offer for every reason, no cancel flow, no exit survey, more than eight reasons, failed payments cancelled with no recovery window, no dunning emails, no pre-dunning, retries that ignore the decline type or are too few, blame in a payment email, poor recovery, a low save rate, no post-cancel path, no reason data, no voluntary / involuntary split, B2B accounts not routed, access ending at cancel, designed dunning emails, no direct card-update link. - Exports: the scan as Markdown, the reason-to-offer map with your savings, the numbers as CSV. ## The paid lanes All run on gpt-terra with a 10% publisher markup; /estimate is free and shows the credits reserved before a run. - audit: the churn picture restated; the leak (voluntary, involuntary, both, unclear) from the page's split; each computed rate read against the benchmarks; the five stages and six dunning steps graded in place, weak, missing or unknown with quotes; cancel reasons getting the wrong offer; risks with severity and quotes; 3 to 6 fixes in order, each naming the lane that builds it; what is missing. - flow: one help-us-improve survey question; 5 to 8 reasons in your data's order, ending with other; one reason-matched primary offer and fallback per reason with the dollar amount from your prices (discounts 20-30% for 2-3 months, never 50%, never past your cap; pauses 1-3 months; no offer for a closed business); the survey, offer, confirmation (access to the end of the billing period) and after-cancel screens, each with a visible way to keep cancelling and no guilt copy; routing high-value B2B accounts to customer success; guardrails; what to measure; one A/B test. - dunning: pre-dunning (card-expiry emails 30, 15 and 7 days ahead, the card updater, a backup payment method); retry plans by decline type (soft: 3-5 retries over 7-10 days; hard: never retried, ask for a new card; authentication: send to update payment); 3 to 5 plain-text emails from day 0 (the skill's day 0, 3, 7, 10) with a direct update link, what they will lose, a support contact and no blame; a grace period and what follows; setup steps for Stripe, Chargebee, Paddle, Recurly or Braintree; recovery targets. - Handoff: an audit hands its leak, fixes, mismatches and risks to the flow or recovery lane in one click. ## How replies are checked The page re-checks every reply against its own scan: the leak against the computed split; metric reads against the page's benchmark reads; no stage graded in place where the notes say it does not exist; high-risk patterns carried as high risks; survey size, order and offers; discount and pause limits; a visible way to keep cancelling; no guilt copy; hard declines never retried; email count, days, update links and no blame; every quote word for word; and no dollar figure or percentage the notes or the page's arithmetic do not contain. ## Sources - Derived from the agent skill @coreyhaines31/churn-prevention by Corey Haines (github.com/coreyhaines31/marketingskills, MIT): https://skillsafe.ai/skill/@coreyhaines31/churn-prevention - its voluntary / involuntary split, cancel-flow structure, exit-survey categories, reason-to-offer mapping, save-offer rules, cancel-flow patterns by business type, dunning stack and email sequence, recovery benchmarks, churn metrics and common mistakes. - Not legal advice: check the cancellation rules (such as the FTC click-to-cancel rule) where you sell.