The Renewal-Risk Score: Reading Churn From Public Signals

Renewals get lost weeks before anyone opens the renewal email. This guide is for founders selling their own product, account executives, SDR leads, and partnerships teams who own a book of live accounts and want to know which ones are drifting toward non-renewal - using only what is visible from outside. No usage dashboard, no support queue, no NPS. Just the people-and-company layer: who left the account, who replaced them, what the org did to itself, and whether the buyer is under financial stress. You will finish able to rank a book into renewal-risk tiers and match each tier to a play.

Almost every published way to score churn reads internal telemetry. That is a problem for two reasons. Account managers and partner teams often cannot see product usage or ticket volume. And telemetry moves late: usage declines about 41% in the quarter before cancellation, which is a warning, but a warning that arrives after the human decision was already made. The external layer moves earlier. This is a framework for scoring that layer.

Why the people layer beats the usage layer on lead time

The public people layer gives you an earlier warning than any usage dashboard, because the decision to leave is made by a person before it ever shows up in the product. A champion's job change is visible the day it posts on a public profile. Usage decline is the echo of a decision that a human already made.

The numbers behind this are stark. When a customer champion leaves, there is a 51% chance the account churns within the next 12 months, according to Sturdy's research reported by ChurnZero, with higher risk when the departing person is a senior leader. Practitioner writeups put the effect at roughly two to three times baseline. Compare that to the usage window: product usage declines by an average of 41% in the quarter before cancellation, which means most churn comes with about a 90-day warning. A champion's exit is visible immediately and carries a heavier signal.

There is a coverage argument too. Telemetry alone reportedly misses around 40% of churn drivers. A budget cut, a reorg, or a competitor-primed new buyer does not always show up as a usage dip until it is too late to act. Reading the people layer catches drivers the dashboard structurally cannot see.

The decision to leave is made by a person before it shows up in a dashboard. Watch the person.

The six public dimensions and what each one proves

Score renewal risk on six public dimensions, ordered by predictive strength. No published study ranks a public-only signal set exactly, so this order is a defensible synthesis anchored on the one hard number available: champion exit at 51%.

Each signal proves something specific, and each has a way it lies. State both, or you will act on noise.

Two rules govern how you read these. First, a single signal misleads: a lone usage dip might be a busy month, and a lone title change might be a promotion. Require two or more independent signals before tiering an account high. Second, silence is a signal, not safety. Roughly 97% of churning customers never contact support. Score no responses to outreach, no data changing on the account, and no business review booked as risk. The quiet account is the one most often mistaken for the healthy one.

Reading the pool: who you can reach and who you cannot

The people layer is not evenly distributed, and that changes what a save play can realistically do. Rebuilding an account after a champion exit is far easier at director and power-user level than at executive-sponsor level, because the senior pool is much thinner.

In Refolk's index of professional profiles, there are far more director-level buyers than VP-level ones. That ratio is the difference between a save play that can work and one that assumes access you do not have.

Director-level contacts outnumber VP-level roughly four to one. This is why the standard multi-threading target, the 1-2-3 rule, weights toward more junior coverage: one executive sponsor, two champions, three power users. After a champion exit you are essentially starting from scratch with the account, and rebuilding down a level is statistically about four times more available than securing a scarce executive sponsor up a level.

The reachable pool also differs sharply by geography, which matters when your book spans markets. Customer Success Managers, the relationship owners you often need to re-thread through, are far more concentrated in the US.

The practical read: a UK-heavy book has a thinner relationship layer to rebuild against, so a champion exit there is proportionally harder to recover from and deserves a higher weight in your score.

The procedure

Run the book through eight steps, front to back. The order matters: exposure before signals, signals before scoring, scoring before thresholds, and confirmation before any play fires.

The scan in step three is where most teams stall, because checking every contact on a book for a job change by hand does not scale past a few dozen accounts. This is the part Refolk removes: you describe the signal in plain English and it returns the matching people and companies across public LinkedIn records and the open web, so a weekly scan of a whole book becomes a query rather than a research project.

Turning scores into tiers, and tiers into plays

Tier by probability crossed with revenue at risk and recoverability, never by score alone. A flagged top-plan account and a flagged small account need different responses, and a champion exit is a different problem from a payment dispute.

The core judgement is two-dimensional. One axis is how strong the public risk signal is. The other is how much ARR is exposed. That produces four responses.

The rule that keeps this honest: internal alerts to your own team should trigger first. Automated customer emails such as product tips or training invitations should only fire for moderate-risk accounts. High-risk accounts need human outreach, not automated emails. An automated discount sent to a healthy high-value account can start a churn conversation that was never happening.

Set the alert threshold from capacity, not statistics. Count how many at-risk accounts your team can genuinely work in a week, and let that number cap your flags. This matters because of base rates. Median B2B SaaS annual churn sits at 3.5% (2.6% voluntary, 0.8% involuntary) per the 2025 Recurly Churn Report, and monthly churn splits by segment.

With logo churn in low single digits to low teens annually, flagging more than roughly 10 to 20 percent of a book as high risk almost always means the threshold is too loose. The discipline is precision, and precision is uncomfortable because 44% of business leaders cannot even state their own company's churn rate. If you do not know your base rate, you cannot tell a good threshold from alarm-flooding.

The tier-matched play pays off. Matching intervention to signal type lifts save rates by around 28% per Totango, and segmenting save offers by tier and severity yields roughly 31% higher net save rate. Detecting decline 45 or more days out saves an estimated 30 to 42 percent of at-risk revenue. The lead time you buy from the people layer is what makes those gains reachable.

How this goes wrong

The failure mode is not missing a churn. It is flagging accounts that were never at risk, firing generic plays, and burning the team's trust in the system. Alert fatigue kills these systems faster than bad signals do. Here is the catalogue and the check for each.

Champion "departure" is a promotion inside the same company. The profile shows a title change, but the person still owns budget. Confirm whether the move is intra-company and whether they retain the account relationship before firing anything.

A neutral new decision-maker gets flagged as risk. Any new buyer triggers red. The real risk is specifically a replacement with a competitor history: if the new buyer used a competitor at their last company, they will often think about swapping you out almost immediately. Check the incoming person's prior stack before you panic.

Layoffs hit the parent, not the using team. A company-wide RIF headline fires the layoff signal, but your buying unit is untouched or even growing. Check which function and location was actually cut.

Silence read as safety. A quiet account looks healthy but has disengaged: no responses to outreach, no data changing, no attendance at the business review. Score absence as risk, not calm.

Threshold set on statistics, not capacity. You generate more flags than the team can work, so none get worked. Check your weekly flag count against your weekly worked-account capacity, and tighten until they match.

Automated save play fires on a healthy high-value account. A false positive damages a renewal that was never at risk. Require human confirmation before any outreach to top-ARR accounts.

Counting a discount-driven bump as a save. Usage or engagement rises after a discount, then the account churns two months later. Apply the 90-day true-save window: a save is an account that returned to a healthy state for at least 90 days after intervention.

Single-signal scoring. Any one signal in isolation misleads. Require two or more independent signals before tiering an account high.

The confirmation gate deserves the most weight, because it is what separates a scoring framework from a nuisance. Sequence it explicitly: predictive scoring identifies who and why, then a human tests that with the customer, and only then does a play fire.

Measuring whether the score works

Judge the framework on two numbers: precision and save rate. Precision is the share of flagged accounts that would have churned without intervention, estimated against a holdout of unworked flags. Save rate is the share of worked accounts that renewed at or above prior ARR.

The holdout is what keeps you honest. If you work every flag, you can never tell whether the score was right or whether you would have renewed anyway. Leave a small set of flagged accounts unworked, watch what happens, and use that as your baseline. This is uncomfortable but it is the only way to estimate precision without internal telemetry.

Apply the 90-day true-save window to every claimed save. A customer who temporarily perks up after a discount and then churns two months later was not saved. Counting them inflates your save rate and hides the false positives. Recalibrate quarterly: adjust dimension weights toward the signals that actually preceded churn in your book, and tighten the threshold if your flagged pool is drifting above real base rates.

Keep the score current

Run this before you call any account's tier final, and re-run the scan weekly so signals surface while there is still time to act.

The book changes every week, so the score is never done. A champion who was safe on Monday can post a new role on Thursday, and a parent company can announce a down round the week before renewal. Set a standing weekly scan against the same six dimensions, feed each confirmed churn back into your dimension weights, and treat the threshold as a dial you tighten whenever the flagged pool drifts above your real base rate. The framework earns its keep in the lead time it buys you: enough weeks between the public signal and the renewal date to do something other than discount.