ChurnSense / Guides

What is churn prediction? A B2B SaaS guide

The short answer

Churn prediction is the practice of using customer behavior and billing data to estimate which accounts are likely to cancel before they do. In B2B SaaS it usually means scoring each customer account for risk from signals such as falling logins, unused seats, support tickets and failed payments, so your team can step in weeks before renewal.

How churn prediction works

  1. Collect data. Product usage (logins, features used, active seats), billing events (failed payments, downgrades), support tickets and satisfaction scores such as NPS.
  2. Turn raw events into measurements. For example logins in the last 7 and 30 days, days since the last login, or the share of paid seats that are active.
  3. Score every account. A rules-based formula or a machine-learning model combines the measurements into one risk score.
  4. Explain the score. A score is only useful if someone can see what drives it, such as "no login in 21 days" or "NPS dropped to 3".
  5. Act. Alert the account owner, contact the customer, fix the underlying problem.
  6. Learn from outcomes. Record which accounts actually cancelled so the next round of scoring is better.

Rules-based health scores vs machine-learning models

ApproachHow it decidesStrengthsLimits
Rules-based health scoreFixed weights you choose, such as "no login in 14 days = red"Transparent, quick to start, no history neededWeights are guesses; they drift as the product changes
Machine-learning modelLearns which patterns preceded past cancellationsAdapts to your product and customersNeeds enough confirmed cancellations to learn from
HybridStarts from a rules-based baseline, then learns from your outcomesUseful on day one and improves over timeMore moving parts to explain

How early can churn be predicted?

Behavior usually changes before a customer cancels: logins taper off, fewer people use the product, support contacts rise. How many weeks of warning you get depends on your product, your contract length and how closely you track usage. A realistic goal for B2B SaaS is to flag at-risk accounts 30 to 60 days before they cancel, and that is the goal ChurnSense is built around. It is a goal, not a guarantee: some customers leave without warning signals.

What churn prediction cannot do

How to start with a small team

  1. Pick five to seven signals you can measure today (see the guide to early churn signals).
  2. Define "churned" precisely, for example subscription cancelled or not renewed.
  3. Rank accounts by risk every week and review the top ten.
  4. Write down what happened to every flagged account. These records become your training data.
  5. Move to a machine-learning model once you have enough confirmed cancellations.

Where ChurnSense fits

ChurnSense is churn prediction software built for small B2B SaaS teams. It scores every account each night, lists the top factors behind each score, and sends a Slack alert when an account crosses your risk threshold. Each customer gets its own model, which starts from a behavioral baseline and switches to learning from your own confirmed cancellations once you have about 20 of them. To understand the numbers behind churn, read how to calculate SaaS churn rate.

Related guides

See which of your accounts are at risk

ChurnSense scores every customer account each night and tells you why. 14-day free trial, no credit card required.

Start free trial