Churn Risk Predictor

Churn is a hazard rate: the share of a starting cohort that terminates within one period. Dividing customers lost by customers at the start of the period gives a discrete period rate, and retention is its complement. Converting between horizons is compounding rather than multiplication, so a monthly rate c implies an annual rate of 1 minus (1 - c) raised to the twelfth power. Five percent monthly is therefore about 46 percent annually, not 60. Under the assumption that the hazard is constant, survival decays geometrically and expected customer lifetime equals 1 divided by the monthly rate, which is why lifetime value is modeled as monthly revenue divided by monthly churn. Benchmarks practitioners use: enterprise B2B software often runs 0.5 to 1 percent monthly logo churn, mid market 1 to 2 percent, self serve SMB 3 to 7 percent, and consumer subscriptions frequently exceed 5 percent per month. The constant hazard assumption is the model's main weakness. Real churn is strongly front loaded, with early cancellations concentrated in the first 60 to 90 days, after which the surviving population is self selected and churns far more slowly. Averaging over a blended base therefore understates long tenure retention and overstates it for new signups, which is why cohort tables and survival curves are preferred for planning. Logo churn also diverges from revenue churn: losing a handful of large accounts can leave customer counts flat while destroying monthly recurring revenue, and expansion within retained accounts can push net revenue retention above 100 percent even when logo churn is material. Definitional choices matter as much as the arithmetic, particularly whether customers acquired mid period appear in the denominator and whether involuntary churn from failed payments is counted separately.

Churn as a managed metric came out of telecommunications rather than software. After US and European mobile markets deregulated through the late 1980s and 1990s, operators competing for a saturating subscriber base began tracking monthly subscriber disconnection rates closely, and churn management became a standard operator discipline. The economic argument was sharpened by Frederick Reichheld and W. Earl Sasser Jr. in their 1990 Harvard Business Review article Zero Defections: Quality Comes to Services, which argued that small improvements in retention produce disproportionately large profit gains, and by Reichheld's subsequent work on loyalty economics. The statistical machinery is older and borrowed from actuarial science and biostatistics. Life table methods for tracking a cohort through successive periods date to seventeenth century mortality analysis, and Edward Kaplan and Paul Meier formalized the modern nonparametric survival estimator in a 1958 Journal of the American Statistical Association paper. David Cox added the proportional hazards regression model in 1972, which allows covariates such as usage, support contacts, and contract age to shift an individual account's hazard rate. Subscription software adopted this toolkit during the 2000s and 2010s as monthly recurring revenue became the standard reporting unit. The period rate computed here is the cohort-level ancestor of that machinery: account-level risk scoring, which maps usage, support contacts and tenure through a regression to an individual churn probability, is the same idea applied to single accounts rather than to a whole starting base.