Churn is the rate at which the machine leaks. It comes in flavours you must never mix:
- Logo (customer) churn = customers lost in period ÷ customers at start. It counts heads.
- Revenue churn = revenue lost from those departures (plus downgrades) ÷ starting revenue. It counts money. When small customers leave and large ones stay, revenue churn < logo churn. That's a good divergence, because the leak is at the shallow end. The reverse, big accounts leaving, is one of the earliest hard signals of product trouble, and you can see it quarters before growth slows.
- Gross vs net. GRR (gross revenue retention) = (starting revenue − churn − downgrades) ÷ starting revenue. It's capped at 100% by construction, because it measures only the leak. NRR (net revenue retention) = (starting revenue − churn − downgrades + expansion within existing customers) ÷ starting revenue. It can exceed 100% when upsell outruns the leak. Both leave out new customers, by definition. NRR ≥ 120% is elite SaaS and < 90% is a leaky bucket. The full treatment, abuse included, belongs to the SaaS metrics page. Here you need the mechanics cold, because LTV runs on them.
Advertisement
How do you convert churn and find customer lifetime?
The conversion arithmetic (drill it until it's automatic). Rates compound, so you can't just multiply by 12. With monthly churn c_m, annual churn = 1 − (1 − c_m)^12. So 3%/month isn't 36%/year. It's 1 − 0.97^12 = 30.6%. Going the other way, 26% annual churn gives monthly churn of 1 − 0.74^(1/12) = 2.48%. Sloppy decks multiply their monthly churn by 12, which overstates it. Or they divide their annual churn by 12, which understates the monthly rate. Either error flows straight into LTV.
Churn-derived lifetime. If a constant fraction c of remaining customers leaves each period, expected customer lifetime = 1/c periods. Why? Survival follows a geometric decay. Start with 100 customers at 20% annual churn. 80 remain after year 1, 64 after year 2, 51.2 after year 3 … and the areas sum to 100/0.2 = 500 customer-years, which is 5 years each on average.
Why can a blended churn rate mislead?
Two health warnings before you ever use it. (1) Real retention curves aren't geometric. They fall steeply early and then flatten. That flattening is the habit floor: customers who survive year 2 are much stickier than year-1 customers. So a blended churn rate computed over a young, fast-growing base is dominated by fragile new cohorts, and it understates the lifetime of the mature base. And a young company projecting its early, floor-less data has no basis for any lifetime claim at all. (2) So take churn from mature cohorts and leave the blended average alone. That takes a retention triangle, which you'll build in cohort analysis. 1/churn is a summary of a curve, so insist on seeing the curve.