In practice
You start March with 200 paying accounts and lose 8 by the end of the month. Monthly churn is 8 ÷ 200 = 4%. You also signed 15 new accounts, but they don't count in either number.
Losing 20 small accounts is different from losing one large one, depending on which measure you use. Churn isn't the opposite of growth: you can add customers faster than you lose them and still have a churn problem underneath.
How to measure it
- Customer churn rate = (customers lost during the period ÷ customers at the start of the period) × 100.
- Gross revenue churn = (downgrade MRR + canceled MRR) ÷ MRR at the start × 100.
- Net revenue churn subtracts expansion revenue, so it can go below zero.
- Pick one period (monthly or annual) and keep it.
Common mistakes
- Including new sign-ups in the numerator or the denominator.
- Watching only customer churn. Chargebee's example: 10% customer churn alongside 30% revenue churn.
- Treating the cancel date as the moment of the decision. Customers usually decide much earlier.
What changes in AI-first products
Churn is late evidence. Agents can ship a fix in a day, but only if someone knows what's driving people out. Annsa has no usage data; it keeps what customers said, and it ranks each problem by the companies and revenue behind it, so a complaint from an account worth keeping is easy to see. Your agents read that over MCP. You decide what to fix.
See how Annsa does it: Customers →