The situation

A subscription business with strong top-of-funnel growth and a quiet crisis underneath it: roughly 65% of customers were gone within a year. Leadership experienced this as a marketing problem, "we need more volume to outrun the leaky bucket", and marketing spend was climbing to compensate. At that churn rate, the unit economics could never work: customer acquisition cost was being amortized over a lifetime too short to repay it.

Diagnose: churn wasn't one problem, it was three

The first move was to stop treating "churn" as a single number. Cohort analysis broke the 65% into three distinct failures with three distinct causes:

  • Onboarding churn (first 30 days): customers who never reached the product's first value moment. The largest bucket, and invisible in monthly averages.
  • Fit churn (days 30–120): customers the marketing was attracting who were never going to succeed, a targeting problem wearing a retention costume.
  • Fatigue churn (120+ days): genuinely satisfied customers drifting away from a product that had stopped evolving with their needs.
One churn number, three problems The 65 points of annual churn, decomposed by cause. Each has a different owner and a different fix.
Annual churn decomposed into three causes Of 65 points of annual churn, 34 came from onboarding in the first 30 days, 19 from poor fit between days 30 and 120, and 12 from fatigue after 120 days. OnboardingFirst 30 days34 ptsFitDays 30 to 12019 ptsFatigue120 days and beyond12 pts

Over half the churn was decided in the first month, before any retention programme engaged. That is why budget moved to onboarding rather than to win-back.

Customer interviews and support-ticket analysis confirmed the split. The critical insight: over half the churn was decided in the first month, before "retention" programs even engaged.

Decide: retention is the growth strategy

The leadership team made three explicit choices, including a painful one. First, reallocate meaningful budget from acquisition to onboarding, accepting slower top-line growth for two quarters. Second, narrow acquisition targeting to segments whose 12-month retention supported the economics, deliberately shrinking the addressable funnel. Third, kill a planned feature expansion and redirect that capacity to the first-value experience.

Design: a 90-day retention rebuild

The one-page strategy centered on a single metric: percentage of new customers reaching the first value moment within 14 days. The roadmap sequenced an onboarding redesign, a segment-level targeting reset with the marketing team, and a win-back program for fatigue-churned customers, the cheapest revenue available, since these customers already understood the product.

Drive: the weekly number that ran the company

A weekly operating cadence tracked cohort retention curves, not the vanity monthly-churn average. When the 14-day activation number moved, the churn number followed it, with a 60-day lag, exactly as the cohort math predicted. Within three quarters, annual churn stabilized at 32%.

Retention by month of tenure Share of an acquisition cohort still subscribed. Cohorts before and after the engagement.
Cohort retention curves before and after the engagement Before the engagement, retention fell to 35 percent by month twelve. After, it held at 68 percent. Most of the difference is established in the first three months. 0%25%50%75%100% M0M3M6M9M12 35% 68%
Before the engagement After
View as table
MonthBeforeAfter
M0100%100%
M172%88%
M262%83%
M355%79%
M450%76%
M547%74%
M644%73%
M742%71%
M840%70%
M938%69%
M1037%69%
M1136%68%
M1235%68%

What moved

  • Annual churn halved: 65% → 32%, roughly doubling average customer lifetime
  • CAC payback moved from mathematically impossible to inside 12 months
  • Marketing efficiency improved as spend concentrated on segments that stayed

The founder-transferable lesson: if your churn number is one number, you don't understand your churn yet. Decompose it by cohort and by cause before spending a dollar fixing it, the fixes for onboarding, fit, and fatigue churn are entirely different, and the averages hide all three.

Disclosure: This describes a real engagement. Identifying details (sector specifics, company scale, and timeline) have been altered to protect client confidentiality. Results figures are as achieved.