How to Predict SaaS Churn — 8 Signals and a Simple Scoring Model

Churn prediction is the practice of identifying accounts likely to cancel before they actually do — giving CS teams enough time to intervene. Most companies attempt churn prediction too late (30 days before renewal), using too little data (just product usage), and without a structured response protocol (just "reach out more"). This guide covers the signals that actually predict churn, when to track them, and what to do when they fire.

When Churn Actually Starts

Churn is almost never a sudden decision. Paddle's SaaS metrics research shows that product usage declines an average of 41% in the quarter before cancellation — meaning the signal appears 90+ days before the customer tells you they are leaving. By the time they say it, the decision is already made.

This means churn prediction is not about catching people who are about to cancel. It is about catching the accounts that are quietly disengaging — before the disengagement becomes a decision.

The Eight Signals That Predict Churn

Not all churn signals are equal. Some are leading indicators — they appear early and give you time to act. Others are lagging — they confirm churn is likely but leave little time to prevent it. Build your prediction model around the leading indicators.

Leading Signals (90+ days before churn)

1. Onboarding completion below 60% at day 30. The strongest single predictor of early churn. An account that is less than 60% through onboarding at the 30-day mark has a dramatically higher probability of cancelling before their first renewal. See our analysis in why SaaS customers churn during onboarding.

2. Days since last client action exceeds 14. When a client has not completed a task, sent a message, or logged into the product in 14+ days, the account is drifting. This signal fires early enough that intervention can still recover the momentum.

3. First value milestone not reached within 30 days. Customers who have not experienced a meaningful outcome from the product in their first month are significantly more likely to churn. Track the TTV milestone and flag any account that exceeds the target date. See our full time to value guide.

4. Champion goes quiet or leaves. When the internal champion stops responding to CSM outreach or exits the company, the account's renewal is immediately at elevated risk. This signal can appear 6–12 months before the renewal conversation.

Mid-Stage Signals (30–90 days before churn)

5. Product usage declining for 6+ consecutive weeks. A sustained downward trend in product usage — not a single dip — indicates the product is being deprioritised internally. This is distinct from a one-week drop due to holidays or a busy period.

6. Support ticket volume increasing. An account that submits significantly more support tickets than average is experiencing friction. Friction, unresolved, becomes a cancellation reason. Track the support-to-value ratio: is the customer getting more problems or more outcomes from the product?

7. NPS drops 20+ points from previous measurement. A significant NPS decline is a direct signal of declining sentiment. More useful than the absolute score is the trajectory — a customer who was at 8 and is now at 4 is in far more danger than a customer who has consistently been at 5.

Late Signals (Under 30 days before churn)

8. No engagement with renewal prep outreach. When a customer does not respond to 90-day renewal prep outreach, does not attend the renewal call, or delays the renewal conversation repeatedly — churn is likely already decided. Intervention at this stage requires executive escalation, not standard CS outreach.

Building a Simple Churn Prediction Model

You do not need a data science team to build a churn prediction model. A weighted scoring system built in a spreadsheet or your CS platform can identify at-risk accounts reliably if the signals are tracked consistently.

A simple model assigns points to each signal based on its predictive strength:

Total score over 40: flag for immediate CSM review. Over 60: escalate to CS manager. This is a starting point — calibrate the weights against your actual churn data over time. For the broader metrics framework this model sits inside, see our post on the customer health score formula.

Response Protocols by Signal Type

A churn prediction model is only useful if it triggers a standardised response. Without a protocol, CSMs receive the alert, feel vaguely concerned, and send a "just checking in" email that accomplishes nothing.

For onboarding signals: see our guide on reducing SaaS churn in the first 90 days for the specific intervention plays.

For mid-stage usage decline: schedule a "success reset" call with the client. Not a check-in. A structured conversation that revisits the original success outcomes, identifies what has not been achieved, and creates a specific 30-day plan to close the gap.

For late-stage renewal silence: escalate to the CS leader, who contacts the economic buyer directly. At this stage, CSM-level outreach has already failed. Executive-to-executive contact is the only lever with a meaningful chance of recovery.