Before They LeaveArrows or space to advance · F for fullscreen · ⌘P to save as PDFSessions

Growth Loops · Session 05 · Dates TBD

Before They Leave

Spot churn earlier and earn a useful return

Ship AIPhoenix, ArizonaPress → to begin

Ship AIBefore They Leave01/19

A cancellation, a failed payment, and fading use are three different problems.

Treating them as one win-back campaign hides the cause and spends customer attention where the product has nothing useful to say.

Ship AIBefore They Leave02/19

Tonight

  1. 01Define churn, the entity, denominator, and time window
  2. 02Separate voluntary, involuntary, activity, logo, and revenue churn
  3. 03Classify activation, fit, service, price, and value-decay causes
  4. 04Find leading signals without turning noise into a risk score
  5. 05Build an at-risk cohort with entry and exit rules
  6. 06Design a value-based intervention and suppression rules
  7. 07Measure durable return and incremental lift against a holdout
Ship AIBefore They Leave03/19

Where we are

  1. Session 01Instrument the Journey
  2. Session 02First Value
  3. Session 03Repeat Value
  4. Session 04Read the Curve
  5. Session 05Before They Leave
  6. Session 06Close the Loop

Retention showed where repeat value decays. Tonight turns that pattern into a precise risk state and an intervention that helps only when there is unfinished value to resume.

Ship AIBefore They Leave04/19

Act one

Define the loss

Churn is a boundary around an entity, an eligible population, an event or absence, and a time interval. The rate is meaningless without all four.

Ship AIBefore They Leave05/19

Name the churn you are calculating

Loss eventDenominator
Logo churnA customer account endsAccounts active at period start
User churnAn eligible user becomes inactiveEligible active users
Activity churnThe value event stops recurringEntities expected to repeat it
Revenue churnRecurring revenue contractsRecurring revenue at period start
InvoluntaryPayment or billing failureRenewals attempted

Expansion and contraction can make revenue move differently from logos. Keep the measures separate before combining them into a net view.

Ship AIBefore They Leave06/19

Work backward from distinct causes

  1. Never activatedThe customer never reached a credible first value.
  2. Value decayedThe job recurred, but the product stopped earning the return.
  3. Poor fitThe customer or job was never a durable match.
  4. Service failureReliability, support, or trust broke the relationship.
  5. Price or packagingThe value and commercial model no longer align.
  6. Payment failureIntent may remain; the transaction failed.

A behavior pattern can support several causes. Pair events with support, billing, interview, and cancellation evidence before choosing an intervention.

Ship AIBefore They Leave07/19

Exercise: separate the causes

~/your-product

classify last 20 losses by evidence

8 value-decay candidates5 never activated4 payment failures3 unknown — research, not campaign

Sample recent churned customers and attach the strongest available behavioral and qualitative evidence. Keep unknown as a valid category rather than forcing every loss into a convenient story.

The counts are your own; the terminal output is illustrative. Small samples guide investigation, not population claims.

Ship AIBefore They Leave08/19

Act two

Build an actionable risk state

A useful signal changes before churn, represents loss of value, and maps to something the product or team can responsibly do.

Ship AIBefore They Leave09/19

Leading signal versus convenient noise

Worth investigating

  • A recurring value action stops
  • Success becomes error or retry
  • A key collaborator disengages
  • Unfinished work accumulates
  • The product's saved state goes stale

Often noisy alone

  • One missed login
  • A lower session count
  • No click on a newsletter
  • A seasonal lull
  • A role that was never expected to act

Compare the same lifecycle stage, segment, and opportunity to act. A signal without eligibility turns normal behavior into false risk.

Ship AIBefore They Leave10/19

An at-risk cohort is a state machine

  1. EligibleThe customer has the job, access, and prior value needed for the signal to matter.
  2. EnterA specific leading condition persists for a defined interval.
  3. ActThe product or team can offer a relevant next-value path.
  4. Exit healthyThe customer completes the value event again.
  5. Exit suppressThey churn, opt out, become ineligible, or enter another owned process.

Without exit rules, the same customer remains at risk forever and receives interventions long after they became irrelevant.

Ship AIBefore They Leave11/19

Match the response to the cause

Useful responseDo not substitute
Never activatedRepair the first-value pathA discount
Value decayResume the unfinished jobA generic win-back blast
Poor fitQualify, export, or offboard wellMore reminders
Service failureAcknowledge, repair, and restore trustFeature education
Payment failureDunning and payment recoveryProduct-engagement messaging

Ownership matters. Billing, support, product, and lifecycle teams should not all contact the same person for the same risk moment.

Ship AIBefore They Leave12/19

Act three

Earn the return

Re-engagement succeeds when the customer resumes useful behavior and sustains it — not when a message earns a click.

Ship AIBefore They Leave13/19

A value-based re-engagement path

  1. TriggerA validated at-risk state, not inactivity alone.
  2. ContextThe unfinished report, task, collaborator, or saved state.
  3. SurfaceIn product, email, human outreach, or billing — matched to the cause.
  4. ReturnThe customer completes the value event again.
  5. RepeatThe behavior recurs after the immediate return window.
  6. StopExit, suppression, consent, and frequency rules end the flow.

Clicks and opens diagnose delivery. The primary product outcome is a return to value that lasts beyond the message.

Ship AIBefore They Leave14/19

Suppression is part of the product

  • No unfinished valueDo not contact customers whose job is complete or no longer relevant.
  • Another process owns itSuppress when support, sales, billing, or an incident response is active.
  • Consent or preferenceHonor channel permission, opt-out, quiet periods, and communication frequency.
  • Potential harmSuppress sensitive, disputed, abusive, or otherwise inappropriate cases for automation.

The right intervention for some customers can be the wrong intervention for the same metric state in another context.

Ship AIBefore They Leave15/19

A holdout reveals incremental return

TreatmentHoldout
EligibilitySame risk and suppression rulesSame risk and suppression rules
ExperienceReceives the interventionDoes not receive it
PrimaryDurable return to valueNatural durable return
Incremental liftTreatment minus holdoutThe counterfactual baseline
GuardrailsComplaints, opt-outs, errorsSame observation window

Customers can return without outreach. A holdout estimates how much return the intervention added beyond what would have happened anyway.

Ship AIBefore They Leave16/19

The artifact

~/your-product

build risk cohort → trigger → holdout → verify exits

entry + healthy/suppress exits validated10% eligible holdout assigned before sendreturn + repeat-value events firingwrote 05-reengagement/incremental-readout.md

Commit the churn definition, cause evidence, risk cohort, intervention, suppression rules, holdout assignment, and incremental-lift scorecard.

Session six uses the strongest proven behavior from the whole lifecycle to build a loop whose output can feed another measurable turn.

Ship AIBefore They Leave17/19

A return is earned when value resumes.

Ship AIBefore They Leave18/19

Next

Close the Loop — Session 06, Workuity Biltmore

The final session maps one product mechanism from input to output and re-entry, finds its constraint, and ships a bounded experiment with an explicit decision rule.

shipai.club · Free, always

Session page
Ship AIBefore They Leave19/19
01/19