Login frequency, NPS drift, champion role changes, support tone, billing failures, competitor mentions. The six signals that surface SaaS churn risk weeks before cancellation, and how to act on each.
Short answer
What are the signs a SaaS customer is at risk of churning?
Track six signals against each customer's own baseline, not an absolute threshold: login frequency drift, NPS drift, champion role changes, support ticket sentiment, billing failures and downgrades, and competitor mentions. The prediction lives in the delta. A 40% drop from an account's normal login rate predicts churn; a low login rate on its own does not.
Most SaaS teams find out a customer is leaving the day they email to cancel. By that point the decision is already three to six weeks old - a champion left, a renewal review went poorly, a competitor shipped something that closed a gap. The save conversation isn't a save conversation; it's damage control on a decision that's already settled.
That gap - between the moment a customer mentally checks out and the moment they tell you - is the entire opportunity space for churn prediction. Six signals show up consistently in the weeks before the cancellation email. They are also, exactly, what Exeechain scores: five of them are components of the scorecard (login frequency, payment health, NPS, champion health, support tickets) and the sixth, a competitor mention, moves champion health. So what follows is the model we run, not a survey of the field. Catch them and the conversation shifts from damage control to relationship repair. Each is below, with what to watch for and how to act.
The single most reliable signal of churn risk is also the simplest: how often is the customer's primary user logging in compared to their own baseline? A 30-day rolling average of daily active sessions tells you, per customer, whether engagement is normal, soft, or collapsed. The absolute number doesn't matter - what matters is the delta from their own baseline. A customer that used to log in nine times a week and now logs in twice has lost something you should know about, even if their plan is enterprise and the MRR is fine.
Action: when login frequency drops 40%+ from baseline over fourteen days, fire a save playbook. The drafted email should reference the specific feature or workflow they used to use most.
A customer who scored you a 9 last quarter and a 6 this quarter has told you, on a 0–10 scale, that something broke in the relationship. That signal is worth more than its absolute score - a 6 is fine if they're always a 6, and disastrous if they used to be a 9. NPS deltas of two or more points downward in either direction are a forced check-in.
The text comments are usually more useful than the score. NPS comments are the closest thing to a free-text exit survey that customers will actually fill in.
The single biggest predictor of enterprise SaaS churn is the champion leaving the customer's organization or moving to a different role internally. The new person inherits a tool they didn't buy and didn't pick - and inherited tools get re-evaluated at the next renewal cycle by default. LinkedIn job-title changes scrape cleanly; HRIS integrations are even better. When the champion changes, the renewal motion starts that week, not at renewal.
Support volume going up isn't the signal - supporttone shifting is. A customer that used to file polite tickets now files terse ones. A customer that asked questions now escalates. Modern LLMs read sentiment well enough that you can score every ticket and watch the rolling average per customer. A 0.4-point tone drop on a five-point scale, sustained for two weeks, predicts churn at 6-8x the base rate.
A customer that's decided to leave often stops caring about the billing relationship before the cancellation email arrives - failed card updates, ignored dunning emails, requests for monthly instead of annual. These signals are obvious in retrospect and easy to miss in real time, because finance and CS systems usually live in different tools. Stripe webhook events into your CS platform fix this for free.
Customers who mention a competitor by name in support tickets, on calls, or in NPS comments are evaluating their options. The most predictive subset is mentions paired with a question (“does your product also do X like Competitor Y?”) - that's a customer comparing feature-by-feature. Catching these requires reading the actual conversation surface, not just metadata.
You predict SaaS customer churn by tracking six signals against each customer's own baseline rather than against an absolute threshold: login frequency drift, NPS movement, champion role changes, support ticket sentiment, billing failures and downgrades, and competitor mentions. The prediction comes from the delta, not the level. A customer logging in twice a week is fine if they always did and is a warning if they used to log in nine times.
The earliest reliable warning signs are a 40% or greater drop in login frequency from that account's own 30-day baseline, an NPS score falling two or more points from the same respondent's previous answer, the departure or role change of the person who championed the purchase, support tickets shifting in tone from curious to transactional, and a failed payment that nobody followed up on.
Usable churn signals typically appear one to four weeks before a cancellation email, not months. Claims of thirty-days-early detection are marketing. The realistic window is the gap between the moment a customer mentally decides and the moment they tell you, and the signals above narrow that gap enough to turn damage control into a save conversation.
Login frequency drift is the single most reliable churn signal, because it is the only one that updates daily, requires no survey response, and is available for every account rather than the minority who answer an NPS prompt or file a ticket. Its weakness is that it says nothing about why, which is what the other five signals supply.
You can, but the prediction is much weaker. Billing records alone show failed payments, downgrades, and non-renewals, which are late signals that appear after the decision is made. Support and survey data add intent. Product usage is what makes the prediction early, so a churn model built on billing data alone is closer to a churn report than a churn prediction.
A fired churn signal should open a specific, dated action rather than a dashboard flag. That means a drafted outreach referencing the exact workflow the customer stopped using, a named owner, and a deadline inside the save window. A signal that produces only a red badge on a screen changes nothing, which is why most health-score projects get abandoned within a quarter.
A signal alone isn't actionable; six signals weighted into a health score are. The pattern: each signal contributes a weighted factor, the score updates daily, and the top three drivers per customer are surfaced in plain English. CSMs read three sentences, know the situation, and have a drafted save email waiting. The fight against churn becomes a fight you can run before the customer has decided.
Exeechain implements all six signals natively. Connect Stripe and drop in a tracking snippet, and the first scores land in fifteen minutes. Connect HubSpot or Salesforce, Intercom or Zendesk, and the model gets sharper week over week. Read the churn prediction feature for the technical details, or the health score guide for how the six signals combine.
Keep reading
Plans & pricing
14 min read · Jul 22, 2026
Getting started
19 min read · Jul 22, 2026
Automation
9 min read · Jul 16, 2026
First scores in 15 minutes. Full accuracy in 24 hours. From $299/mo. Daily AI churn scoring with the top three drivers per customer.
Not ready to switch? Run the free 60-second revenue leak scan and see what staying put costs.