Every subscription ends twice: first the customer stops getting value, then they cancel. The cancellation email is the second event. This lesson works backward from it — distinguishing why people leave, building a loop that learns from each exit, and refusing to let reminders or discounts cover for a product that stopped helping. (Product track: the same exit-question discipline returns for member exits in Lessons 83.2 and 83.5.)

Three kinds of leaving

• Voluntary churn: the customer chooses to end or not renew. The briefs stopped mattering, the team changed tools, the budget moved, the workspace never became a habit. This is product evidence — read it as a verdict on value.
• Failed-payment churn: the customer did not choose to leave; the card expired, the bank declined, the invoice went to someone on leave. This is operations evidence. Dunning emails, card-update flows, and retry logic from your billing provider recover much of it — see the Stripe Billing documentation for retry and recovery mechanics — but none of it tells you whether the product was loved.
• Inactive-but-still-paying: the quietest and most dangerous kind. The card works, the login does not. These customers look like revenue and behave like departures — until the day they notice the bill and leave angry, often with a chargeback and a story. Track active use of the core job, not just active billing.

Conflating the three corrupts every decision. Celebrating "low churn" while failed payments silently resolve and inactive accounts pile up is dashboard decoration. Split them before you interpret anything.

Retention first means value first

The best retention work begins with a product that delivers recurring value. Reminders and discounts cannot repair a product that no longer helps — they can only bribe a customer to notice the gap once more.

Order the work honestly. First, the core job happens on rhythm (Lesson 81.1's recurring value). Second, activation gets the customer to first value fast (Lesson 81.2's door). Third, the billing unit stays legible so renewal never surprises (Lesson 81.3's unit). Only then do save tactics earn their place: a well-timed "your watchlist caught 3 changes this month" summary before renewal, a pause-instead-of-cancel option for seasonal users, a humane downgrade path that preserves exports.

Discounts deserve special suspicion. A discount that saves a customer who never activated teaches you to discount instead of onboard. Offer the exit interview before the coupon: "What would have to be true for this to be worth full price?" The answer is product roadmap; the coupon is fog.

The cancellation-learning loop

Treat every cancellation as a six-step investigation, not a metric tick:

1. Reason: capture what the customer says, in their words, at exit — one respectful question, optional, never a gauntlet. 2. Evidence: preserve what the product saw — last active date, core-job usage, plan, tenure, support history. Memory fades; logs do not. 3. Categorize: file it into one of five actionable buckets (below) — never "other" without a note. 4. Cohort: inspect the pattern — do trial converts leave in week two? Do annuals vanish at month eleven? One exit is anecdote; a cohort is instruction. 5. Change: alter exactly one thing — onboarding, allowance, brief rhythm, paywall clarity — tied to the category. 6. Verify: watch the next cohort. Did the change move the category's rate? If not, the diagnosis was wrong, not the data.

Loop monthly while small; the sample is tiny, so read every exit like a letter, not a percentage.

The simple retention table

Hold four columns on one page. Complexity can wait; visibility cannot.

A customer who never reached first value and cancels is an onboarding failure, not a pricing failure. A customer active for five months who leaves after the brief rhythm slipped is a delivery failure. The table tells you which room to enter.

Exercise: design the churn learning log

Create CHURN-LEARNING-LOG.md with a respectful exit question and five categories the team can actually act on.

# CHURN-LEARNING-LOG.md — [Project]

## Exit question (asked once, optional, at cancellation)
- "What would have to be true for [product] to be worth keeping? (one line is plenty — and your exports stay available for 30 days)"
- Asked where: ___ / who reads replies: ___ / reply SLA: ___

## Five actionable categories
1. Never activated (no first value) → owner: onboarding — fix: ___
2. Value faded (used, then stopped) → owner: product/rhythm — fix: ___
3. Wrong fit (job or buyer mismatch) → owner: positioning/paywall — fix: ___
4. Price or unit shock (bill surprise) → owner: pricing/support — fix: ___
5. Failed payment / admin lapse (did not mean to leave) → owner: billing ops — fix: ___

## Log (one row per exit)

| Date | Tenure/plan | Last active + core-job use | Stated reason (their words) | Category (1–5) | Cohort note | Change queued | Verified? |
| --- | --- | --- | --- | --- | --- | --- | --- |
| ___ | ___ | ___ | "___" | _ | ___ | ___ | ___ |

## Monthly review (30 minutes)
- Biggest category this month: ___ / cohort pattern: ___ / one change shipped: ___ / next check date: ___

Worked mini-example — Desk: exits cluster in category 1 in the first month ("never got my first brief set up") and category 2 around month four ("briefs feel repetitive"). The loop prescribes exactly two changes: a 48-hour activation checklist with a sample topic preloaded, and a brief-variety pass that rotates sources and flags "no meaningful change" honestly instead of padding. Failed payments get their own recovery flow via the provider — see retry and recovery mechanics in the Stripe Billing documentation — and are never counted as product verdicts. Keep plan structure comparable using the Stripe recurring pricing models.

Finish line: a retention system that teaches why people leave — exit question live, five categories owned, one monthly review scheduled.

Verify quickly: file your last three real or rehearsed cancellations. If two land in the same category and you cannot name the change it prescribes, the categories are labels, not levers — rewrite until each maps to an owner and a fix.

Common failure mode: the discount bandage — answering every exit with 20% off. It preserves the revenue line for one more month and destroys the learning line permanently. Ask first, discount last, and only for price-shock exits where the value evidence is strong.

Check your understanding

1. Distinguish voluntary, failed-payment, and inactive-but-paying churn — and why each demands a different response. 2. Why can't reminders and discounts repair a product that stopped delivering recurring value? 3. Recite the six steps of the cancellation-learning loop and what each produces.

Next

Retention is a conversation; economics is its ledger. Lesson 81.5 turns customers, costs, and cohorts into numbers you can actually reason with — without the theater.