Most SaaS businesses learn about churn at the worst possible moment: the cancellation screen. By then the customer has already decided, usually weeks earlier, and the survey they fill in on the way out tells you almost nothing useful. "Too expensive" is rarely about price — it is what people say when the product stopped being worth the line on the invoice.
The useful information was available long before that, in how the account was being used. Reading it is not a machine-learning project. For most products under a few thousand customers, four or five simple signals catch the majority of preventable churn.
Renewal-cycle thinking is the underlying problem. If your only checkpoint is the invoice date, you are looking at customers once a year at exactly the point where they have already built their case for leaving. Usage moves months earlier and moves continuously.
The shift that matters is from asking "who is up for renewal?" to asking "whose behaviour changed this month?" — a question you can answer every week from data you almost certainly already collect.
Start with these. They are boring, cheap to instrument, and they work.
Not total seats — seats that logged in this month versus last. A team account that quietly drops from nine active users to three has already made its decision internally, whatever the admin says on a call.
Every product has one action that represents genuine use: an invoice sent, a report generated, a campaign published. Track that, not logins. People log in to cancel.
Accounts that never completed the setup step that makes the product useful are a churn cohort from the moment they subscribe. Most products can name that step in one sentence; far fewer measure how many customers reach it.
Two things predict departure: a spike in tickets about the same unresolved problem, and a long-standing customer who suddenly stops contacting you at all. Silence from an engaged account is a warning, not a compliment.
Detection without a response is just a sadder dashboard. Each signal needs one predefined action, and the action should match the cause.
Never activated? That is an onboarding failure, and it is fixable with a human. A short offer of setup help, from a real person with a calendar link, recovers more revenue than any automated email sequence.
Usage falling on a healthy account? Something changed on their side — a champion left, a process moved elsewhere. Ask directly, early, without a sales pitch attached.
Repeated support pain? Fix the underlying issue and then tell the customer specifically that you fixed it. Closing that loop converts an irritated user into a reference.
A discount offered at the cancellation screen buys a month. Solving the problem three months earlier buys a renewal.
Whatever you do, do not save the intervention for the cancellation flow. A retention offer at that point trains customers to threaten to leave, and it papers over the reason they wanted to.
Churn prediction has stopped being a differentiator — plenty of tools will score your accounts now. What separates products that keep customers is what happens after the score changes: a specific, human response matched to the reason, delivered while the customer still has a problem worth solving rather than a decision already made. Instrument four signals, review them weekly, and write down what you will do about each one. That is the whole system.
We build the usage tracking, health scoring and in-product flows that catch churn while it is still reversible.
Talk to us