Churn stopped being a renewal-week surprise: warning lead time went from 11 days to 96, and net revenue retention from 97% to 108% over nine months.
Churn was discovered at renewal. Health scoring existed but nothing in it predicted anything, so nobody used it.
Net revenue retention
97%
Expansion less contraction and churn, same cohort basis.
Churn warning lead time
11 days
Median days between first triggered risk signal and the renewal decision.
Health score predictive value
None measurable
A score existed. It did not separate churned accounts from retained ones in back-test.
Onboarding to first value
76 days
Contract start to the first defined value behaviour in product.
Renewals worked in final 30 days
68%
Of renewal value, first substantive conversation inside the last month.
The seam was the nine months between onboarding completion and the first renewal read. Nobody owned the account signal in that window, so churn was discovered rather than prevented.
Preventable churn
$3.4M
Churned ARR where a back-tested behavioural signal had fired more than 60 days before the decision.
Contraction absorbed inside a strong expansion quarter
$1.1M
Not visible in reporting until expansion and contraction were split.
Emergency save discounts
$720K
Discount granted in the final 30 days of a renewal cycle, trailing four quarters.
$5.22M a year lost in a window nobody was accountable for.
01
Month 1
Took 24 months of churned and retained accounts and tested which behaviours actually preceded a loss. Retired every scoring input that failed the test.
What we read
Four behaviours separated the populations. The existing 14-input score did not.
02
Months 2–3
Replaced the onboarding checklist with a completion definition tied to observed product behaviour, and instrumented it.
What we read
31% of accounts marked onboarded had never reached the behavioural definition. That gap became the first work queue.
03
Months 4–6
Built the trigger set, assigned an owner per signal, and moved the first substantive renewal conversation to day 90 with a defined agenda.
What we read
Warning lead time 11 → 62 days. Save-discount usage started falling immediately.
04
Months 7–9
Split expansion from contraction so a strong upsell quarter can no longer hide churn. Ran the weekly signal review with the customer team, then withdrew.
What we read
NRR 108%, warning lead time 96 days, both read on the definitions set in month one.
Net revenue retention
97% → 108%
+11 points; expansion and contraction reported separately from month seven.
Churn warning lead time
11 → 96 days
+85 days of usable warning before the renewal decision.
Onboarding to first value
76 → 41 days
−35 days against a behavioural definition, not a checklist.
Renewals worked in final 30 days
68% → 19%
By renewal value, on the same cohort basis.
The signal set, the day-90 trigger and the split reporting run inside the customer team's own weekly cadence.
This is an illustrative composite. The numbers, phases and outcomes show how we baseline a motion, read a delta and hand the cadence back — they are modelled on the shape of RevOps work delivered through the network, not lifted from one named client. We publish nothing under a client's name without written sign-off.
One large account renewed on a multi-year term inside the window. NRR is reported both with and without it internally; the figure above includes it and is flagged as such on the record.
The customer management in fullWe start the same way every time: one baseline, one written definition, one cadence you keep after we leave.