Churn Prediction for B2B SaaS: The Signals That Fire Before the Renewal

Churn prediction should fire months before the renewal

Most churn prediction is a post-mortem dressed up as a forecast. By the time usage dips and the survey comes back cold, the customer already decided. The earliest reliable churn signals show up somewhere your renewal dashboard is not looking: the support queue and the close-date field. At ORM we see both move weeks or months before the number does.

Renewal risk is not a mystery you solve in the QBR before the contract ends. It is a pattern that accumulates. The teams that catch churn early are not the ones with the fanciest customer health score. They are the ones watching the two behaviors that change first: how a customer uses support, and how their buying behavior drifts once the relationship goes quiet.

The support-ticket U-curve

Here is the finding that surprises most revenue leaders. Support-ticket volume predicts churn, but not in the direction they expect. The relationship is a U-curve, not a line.

At ORM we see it consistently. A customer filing zero support cases is not a happy customer. It is an absent one. Nobody is logging in deep enough to hit friction, nobody is invested enough to complain, and nobody has built a reason to renew. On the other end, a customer filing seven or more cases in a year is drowning. The product is not working for them and they are telling you in the only channel they have.

The healthy middle is the part people get wrong. A customer filing three to five tickets a year, usually tier 2 or tier 3, the ordinary not-severe kind, is your safest account. They are engaged. They are getting help. They are building the product into their workflow. That friction is a sign of investment, not dissatisfaction.

Score support volume as a curve, not as "fewer tickets is better." A quiet queue feels like success to a support manager and reads like a warning to anyone who has watched these accounts churn.

Silence is the earliest signal

Support volume tells you about the account. Behavior change tells you about the moment risk turns real. And the earliest signal of all is not a signal. It is the absence of one.

I learned this reading deal slippage patterns on the new-business side, and it transfers directly to renewals. The strongest predictor that a deal is dying is not a bad meeting or a lost champion. It is silence. No activity, no data changing, no notes, no replies. When a buyer stops returning your emails, stops picking up the call, and stops texting back, the deal is already gone even if it still sits in commit.

Renewals decay the same way. The account that stops opening your check-ins, stops attending the QBR, and stops responding to the CSM has made a decision it has not told you about yet. Layer that on top of the support U-curve and you get a compound signal. An account that went from three tickets a year to zero and stopped answering is not stable. It is pre-churn.

The other behavior worth watching is any change to expansion timing. When a customer keeps pushing the date on an upsell, treat it the way a forecaster treats a rep moving a close date. A slip is a downgrade in probability, not a neutral scheduling update.

The Silent Renewal early-warning system

Here is the framework I use. I call it the Silent Renewal system, and it has three inputs, all of which move before the renewal date.

1. Support shape. Track trailing-twelve-month ticket volume against the U-curve. Flag any account at zero, any account at seven or more, and any account whose volume collapsed year over year. 2. Engagement pulse. Measure response, not merely activity. Emails opened is noise. Emails answered is signal. An account trending toward silence is trending toward churn. 3. Behavior drift. Watch for pushed expansion dates, shrinking usage, and stalled onboarding of new seats. Each is a close-date-style slip applied to an existing customer.

Score all three together, not separately. The power is in the combination. Zero tickets alone might mean a low-touch product fit. Zero tickets plus fading responses plus a pushed expansion is a renewal you are about to lose. Wire these into the same model that runs your revenue forecasting, because churn is a revenue event and belongs in the same forecast as new and expansion, not quarantined in a customer-success tool nobody reconciles against the number.

A worked example (illustrative, not a benchmark)

Take Larkfield Analytics, a fictional mid-market account. The numbers are illustrative, not a benchmark.

Larkfield renews in five months. On paper it looks safe. Paid on time, no complaints, no escalations. The health score is green, because green is what "no complaints" produces in most scoring models.

The Silent Renewal system reads it differently. Larkfield filed four tickets last year and zero in the last seven months. Their CSM's last three check-in emails went unanswered. The expansion they discussed in Q1 keeps sliding a month at a time. Three inputs, all pointing down, none of which shows up as a red health score.

That is the account you call this week, not next quarter. Caught now, a save is a conversation. Caught at the renewal, it is already churned ARR on your retention waterfall.

Why quiet accounts fool everyone

Here is the contrarian position I will defend. Your customer health score is probably scoring the wrong direction on support. Most health models reward low ticket volume and penalize high volume, treating every ticket as pure dissatisfaction. That gets one tail right and the other tail exactly backward. The silent, zero-ticket account scores as your healthiest and is often your most dangerous.

The deeper mistake is treating churn as a customer-success problem measured after the fact by churn rate and gross revenue retention. Those are lagging metrics. They report what already happened. Prediction lives upstream, in the behavioral data that moves first. A model that only learns from closed churn is always a quarter late.

Build the system before you need it

Churn prediction is not a dashboard you buy the quarter before a renewal cliff. It is an early-warning system you build from the signals that move first: the shape of the support queue, the pulse of engagement, and the drift in buying behavior. Watch those three and you get months of lead time instead of a post-mortem.

At ORM we model churn in the same forecast that runs your new and expansion revenue, so a renewal at risk shows up on the number while you still have time to save it.