Product Analytics for SaaS Retention: The Metrics That Actually Predict Churn

The problem with churn as a metric

Churn rate is a lagging indicator — it tells you what happened, not what's about to happen. By the time monthly churn ticks up, the disengagement that caused it started weeks earlier. Optimizing for churn rate directly (e.g., by adding exit friction or offering discounts at cancellation) addresses symptoms. Product analytics lets you work on causes: identifying which usage patterns predict churn and intervening at the behavioral level before the decision is made.

L28: The single best leading indicator of SaaS churn

L28 — days active in the last 28 days — is the most reliable leading indicator of churn we've found across SaaS business models. A customer who was active 20 out of 28 days last month is not a churn risk. A customer who was active 3 out of 28 days is almost certainly disengaging. Track L28 at the account level (not user level, for B2B), and create automated alerts when an account's L28 drops below your empirically-determined threshold. For most SaaS products, a drop below 30-40% of typical usage is a strong churn signal.

Feature adoption breadth

Customers who use only one feature of your product are far more likely to churn than customers who have integrated multiple features into their workflow. Single-feature users can be replaced by a point solution; multi-feature users have switching costs. Track the number of distinct core features each account has triggered in the last 30 days. Accounts with feature adoption breadth below 2-3 features (depending on your product's depth) should be flagged for customer success outreach. The goal: get them to a second feature before they cancel.

Collaborative signals for B2B SaaS

For B2B products, individual user activity isn't enough — you need to track whether the product is embedded in team workflows. The leading indicators here are: number of active users within an account (not just seats purchased), frequency of account-to-account sharing or collaboration events, and whether the account has completed any integration setup (API connections, data imports, SSO). Accounts where only one person is active, even if they're a power user, are at significantly higher churn risk than accounts with distributed activity.

The 'dark period' signal

One of the most predictive churn signals is a sudden drop in activity after a period of normal or high engagement — what we call a dark period. A customer who was active daily and suddenly goes 10+ days without logging in hasn't quietly solved their problem; they're almost certainly evaluating alternatives or have lost their internal champion. Set up an automated alert in Amplitude or your CRM for accounts that trigger 0 events in a rolling 7-day window after averaging 5+ active days per week in the prior 30 days. This signal has the highest urgency — these accounts need a customer success reach-out within 24-48 hours.