How to Find Customers at Risk of Churning Before They Leave

At-risk customers announce themselves weeks before they cancel. The average B2B SaaS customer decides to leave 30 to 60 days before submitting the request. Usage-based scoring surfaces that risk 45 to 60 days out. The window exists. Most teams are not watching it.

The reason is uncomfortable. Around 97% of churning customers never contact support. The loudest accounts are rarely the ones leaving. The quiet ones are. Waiting for a complaint means waiting for a cancellation.

This guide covers the eight signals that predict churn, ranked by how early they appear. It also covers what to do once an account is flagged. For the scoring model that combines these signals, see the companion guide on health scoring.

TL;DR: Finding At-Risk Customers

Why At-Risk Customers Go Undetected

The detection problem is structural rather than a matter of attention, as research on churn warning signals consistently shows. Three things work against you.

That third point is decisive for app businesses. A customer success manager can track forty enterprise accounts by relationship. A Shopify app with two thousand merchants and no CSM team cannot. Automated detection becomes the only workable approach, not a nice upgrade.

The cost of missing the window is measurable. Teams catching decline 45 or more days early recover an estimated 30% to 42% of at-risk revenue. Teams that find out at cancellation recover almost none.

[Image alt text: timeline showing when at-risk customers become detectable before cancellation]

The 8 Churn Warning Signs, Ranked

Not all churn indicators are equal. Some give you 60 days of warning. Others confirm what you already know. These are ranked by how early they appear.

Two entries deserve elaboration. Signal 4 is widely missed. A merchant switching from annual to monthly billing is not a lost customer today, but is telling you about tomorrow. Signal 8 is real but arrives late and covers few accounts. Use sentiment to explain risk, not to detect it.

Reading Signals Correctly

Most detection failures come from measuring the wrong thing rather than measuring nothing.

Measure velocity, not level

A merchant at 40 events per week is uninteresting. One who fell from 40 to 12 over three weeks is the whole point. Churn research points to a sustained drop of 30% or more against the account's own baseline. Three to four weeks is the clearest window.

Baseline per account, not globally

Usage varies enormously between merchants. A global threshold flags heavy users who slowed slightly. It misses light users who stopped entirely. Every account needs its own reference point.

Prefer fast signals over slow ones

Usage changes and payment failures move quickly. NPS surveys and reviews move slowly, arriving after the decision. A system built on slow signals fires too late to matter.

Separate churn types

A merchant leaving at month three failed onboarding. One leaving at month eighteen outgrew you. Same outcome, opposite intervention. Route them differently.

Prioritising Which Accounts to Save

Once detection works, you will have more flagged accounts than capacity. Prioritisation decides whether that list is useful or overwhelming. Four factors rank it.

Weighting by revenue is the step most teams skip. Treating every flagged account equally means spending the same effort on a merchant worth $9 a month as one worth $400. Pairing risk detection with merchant lifetime value turns a flat alert list into a ranked queue.

Save Plays That Work

Detection without intervention changes nothing. Each signal maps to a different response.

Two principles govern all of them. Speed beats polish. Accounts contacted within days of a first warning retain better than those reached after a third. Predictive models driving proactive outreach cut churn by around 10%. That is a large return for a workflow change.

Detecting At-Risk Customers at Scale

For a self-serve app business, nobody is reviewing accounts one by one. Detection has to run automatically or it does not run at all.

Automated risk detection in Elevate works from your Shopify Partner data, so the signals above surface without instrumentation or manual review.

With Elevate, Shopify app teams can identify at-risk customers in the same place they track subscriptions, plan changes, payments, reviews, revenue and merchant activity. Instead of treating every churn signal equally, you can see the context behind each account, understand what changed before churn risk appeared, and prioritise retention efforts based on revenue at risk. This gives self-serve Shopify SaaS teams a more practical way to turn customer churn prediction into timely action. 

[Image alt text: dashboard highlighting at-risk customers ranked by revenue and churn signal strength]

To turn these signals into a single ranked number, see customer health scoring, which covers weighting and scoring bands. For the revenue side of the same picture, revenue churn tracking shows what the accounts you miss are actually costing, and install source data reveals which acquisition channels produce at-risk merchants in the first place.

*Volumes are directional ranges, not a tool export. Validate against your own SEO platform before locking a content plan.

Almost all ranking content assumes a customer success team with named account owners. Sigma writes for data teams building retention dashboards, while ChurnZero, Kapta, and CustomerGauge write for CSMs managing enterprise portfolios. None address a self-serve product with thousands of low-touch accounts and no CSM function, which is exactly the Shopify app situation. That framing difference, not the signal list, is what makes this page distinct.

Reducing customer churn starts with seeing risk early enough to act on it. For Shopify app businesses, that means connecting customer health, subscription analytics, revenue churn, payment events and merchant behaviour instead of reviewing each signal in isolation. Elevate brings that Shopify app analytics data together so teams can find at-risk merchants earlier, understand why they may churn, and focus retention efforts where they can have the greatest impact. 

Frequently Asked Questions

How do I identify at-risk customers?

Track usage against each account's own baseline and watch for core feature abandonment. Monitor failed payments and downgrades, and flag accounts that go silent. Usage signals surface risk 45 to 60 days before cancellation.

What is the earliest sign a customer will churn?

A sustained decline in active usage measured against that account's normal level. A drop of 30% or more over three to four weeks is the clearest pattern. It appears well before any human-visible warning.

Why do most churning customers never complain?

Because disengagement is quiet. Around 97% of churning customers never contact support. They stop logging in and cancel without raising an issue. Silence is a signal, not reassurance.

When is churn risk highest?

In the first 90 days. Between 40% and 60% of SaaS cancellations happen in that window. The usual cause is never reaching the first value. Early lifecycle detection has the highest return.

Does contacting at-risk customers actually reduce churn?

Yes. Detecting decline 45 or more days before renewal saves an estimated 30% to 42% of at-risk revenue. Predictive models driving proactive outreach cut churn by around 10%.

Should I treat all at-risk accounts the same?

No. Rank by revenue at risk and recoverability. A flagged top-plan merchant and a flagged trial user need different responses. Failed payments are far more recoverable than product-fit churn.

Is a plan downgrade a churn signal?

Yes. Downgrades and switches from annual to monthly billing both indicate reduced commitment. The customer stays in your count while signalling that they are keeping their options open.

How is this different from a customer health score?

These are the underlying signals. A health score combines them into one weighted number with scoring bands. Identify which signals matter for your product first, then build the score on top.