The renewal call is sixty days out, and it goes badly in the first five minutes. The customer is consolidating vendors, the new VP of Operations has a preferred tool from her last company, and the decision was made last month. The account had been green all year. When the customer success manager reconstructs the story afterward, every piece was there. The champion who bought the product left in the spring. Weekly active users had been sliding since. Three support tickets in the last quarter used the word "frustrated." The new VP had never once logged in.
Elsewhere in the same book, an account doubled its seat usage, hit its plan limit twice and asked support about a higher-tier feature. Nobody called them either. The expansion happened at renewal, at a discount.
Both stories are the same failure. The information existed. It lived in four systems, belonged to no one and triggered nothing.
The stakes have moved. SaaS Capital's 2025 retention benchmarks, drawn from more than 1,000 private B2B SaaS companies surveyed in early 2025, put median net revenue retention at 101% and median gross retention at 91%. The same report found that companies with NRR above 130% grow at roughly double the population's median rate of 24%, and that the 110% to 120% NRR group had median growth of 30%, nine percentage points above the 21% in the 100% to 110% group. Gross retention alone showed little correlation with growth. Expansion is what separates the fast from the flat.
Boards and buyers know this. Benchmarkit's 2025 SaaS Performance Metrics report (583 participants) found that expansion made up 40% of total new ARR at the median in 2024, up from 25% in 2022, as report partner Maxio summarizes. Gartner's survey of 243 CSOs and senior sales leaders, fielded in late 2024, found 73% prioritizing growth from existing customers for 2025 and 57% naming account retention and growth as a top-three priority. The installed base has become the growth plan.
Yet the instrument most teams use to watch that base is weak. In Hook's Customer Success in 2025 report announcement, CS teams rated the accuracy of their own health scores at an average of 5.96 out of 10, and 30% had no health score at all. Hook's public announcement confirms both figures, but does not establish the sample size. Underneath it, Validity's State of CRM Data Management in 2025, a survey of 602 CRM users and administrators, found 76% saying less than half of their CRM data was accurate and complete. A retention forecast built on that foundation is a guess with a color attached.
Diagnosis: why NRR surprises keep happening
At $3M to $30M ARR, the customer success function usually grew faster than its data. Neither the first health score nor its replacement was designed around the question that matters: what will this account do at renewal, and what should we do about it now? Five patterns explain most surprises.
Health scores built from what is easy to measure
The typical score blends login counts, NPS responses, ticket volume and a CSM's gut-feel rating, each with a weight someone chose in a workshop. Logins say little about whether the product is doing the job it was bought for, and NPS reaches a fraction of users and lags behavior. The score ends up describing activity, not intent, and nobody has ever checked it against which accounts actually churned.
Signals scattered across systems no one joins
Product usage lives in the product database or an analytics tool. Tickets live in the help desk. Contacts and contracts live in the CRM. Billing lives in finance. The decline that predicts churn is rarely visible in any one of them; it is the combination, falling usage plus a frustrated ticket plus a departed admin, that matters. Without an identity layer that ties every event back to one account record, the combination never forms.
Relationship risk that nobody tracks
People change jobs constantly. The U.S. Bureau of Labor Statistics put median tenure with the current employer at 4.1 years for wage and salary workers in January 2026. On a three-year contract, there is a real chance the person who signed it will not be there to renew it. Yet most CRMs record a champion once, at the opportunity, and never verify that the person still works there, still holds the role, or still uses the product.
Expansion treated as a sales event, not a signal
Upsell usually waits for a renewal conversation or a rep's quarterly account review. The usage patterns that precede expansion, such as seat utilization near the limit, new teams or departments adopting the product, or repeated use of features at the edge of the current plan, are often visible in the product months earlier. They are routed nowhere, so expansion arrives late, small and discounted.
Alerts without owners, plays or outcomes
When teams do automate alerts, they tend to fire into a shared channel or a dashboard nobody opens on a Tuesday. There is no named owner, no expected action and no record of what happened next. After a few weeks of noise, the alerts are muted.
The framework: the NRR Early-Warning System
The NRR Early-Warning System replaces the single blended health score with four signal families, two separate scores and a closed loop. It runs inside the CRM, because that is where owners, renewal dates and revenue already live.
Four signal families, each with a lead time. Usage covers depth and breadth of adoption: active users against purchased seats, use of the features tied to the customer's stated outcome, and the trend over 30 and 90 days rather than the level on any one day. Sentiment covers support ticket volume, severity, reopen rates and the language in the tickets themselves, which a classifier can tag as frustrated, blocked or neutral. Relationship covers the people: whether the champion and economic buyer still hold their roles, whether a new executive has arrived, and how many distinct people at the account engage with you. Commercial covers what finance sees: payment delays, downgrade requests, seat reductions and how far away the renewal date is. Each signal is documented with how early it typically fires before a renewal outcome in your own history and how its freshness decays, because a signal that fires two weeks out is a report, not a warning.
Two scores, not one. Risk and expansion are scored separately, because an account can be both at once: a growing team in one department and a departing champion in another. A single blended number averages those into a reassuring yellow. Two scores keep both stories visible and route them to different people.
Every alert has an owner, a play and a deadline. A risk alert on a strategic account goes to the named CSM with a specific play (an executive check-in, a usage review, a re-onboarding session for the new admin) and a date by which it must be logged. An expansion alert goes to the account owner with the evidence attached. If no one acts, the alert escalates rather than expiring quietly.
Outcomes close the loop. Every renewal, downgrade, churn and expansion is written back against the signals that preceded it. Once a quarter, the weights are rechecked against what actually happened. Signals that never predicted anything are removed; signals that did are given more weight.
Implementation: six steps to an early-warning system you can trust
You do not need a new platform to start. You need your renewal history, one account record and the discipline to test first.
Reconstruct two years of renewal outcomes
Pull every renewal, churn, downgrade and expansion from the last eight quarters, with the dollar change for each. Then, for each account, look back 90 to 180 days and record what the usage, ticket, contact and billing data showed at the time. The diagnose-before-you-build playbook covers how to do this read-only. Check: you can list your churned and expanded accounts with the signals that preceded each.
Join the signals to one account record
Map product workspaces, help desk organizations and billing customers to CRM accounts, and resolve the duplicates. Without this, usage from one workspace and tickets from another never meet. Check: at least nine in ten active product workspaces and support organizations resolve to a single CRM account, as a suggested starting point, not a benchmark.
Define the signal taxonomy and lead times
For each of the four families, write down the specific signals, the threshold that counts as a change, and the lead time your history shows. A starting set to adapt: weekly active users down 30% over 60 days, two or more frustrated tickets in a quarter, champion or economic buyer departed, payment more than 30 days late; and on the expansion side, seats above 85% of plan, a new department active, or repeated attempts to use higher-tier features. These are suggested starting points, not benchmarks. Check: each signal has a definition, a source system and a measured lead time.
Build the two scores and backtest them
Combine the signals into a risk score and an expansion score, then run both against the history from step one. The account-scoring backtest explains why weights must come from outcomes rather than a workshop. We hold every system to the same bar: tested on around 20 of the client's own past cases, and 85 percent correct or it does not ship. Check: the scores flag most of last year's churned and expanded accounts early enough to act, and every miss has a written reason.
Design the alert routing and plays
Decide who receives each alert by segment and account value, what play they run, and how long they have to log it. Keep weekly alerts per CSM few enough to act on. Check: every alert type has an owner, a play, a deadline and an escalation path.
Run in shadow, then switch on and review quarterly
Generate alerts for a few weeks without sending them, and compare them with what CSMs already know. When the shadow alerts hold up, turn them on, log every play and outcome, and review the weights each quarter against actual renewals. Check: leadership signs off on the shadow results before alerts go live.
Workflow: the early-warning loop
A suggested operating loop. Adapt the cadence to your renewal cycle; keep the owners.
What happens: product usage, support tickets, contact and role changes, and billing events arrive daily and are matched to one CRM account.
System role: make every signal about an account visible in one place, so a usage drop and a frustrated ticket are seen together rather than in separate tools.
Owner: RevOps owns the data connections and account matching.
What happens: each account's risk and expansion scores are recalculated, and any account that crosses a threshold or moves sharply is flagged with the specific signals that moved it.
System role: turn many small changes into one clear, explainable reason to act, using the same written rules for every account.
Owner: RevOps maintains the model; the Head of Customer Success approves changes to weights and thresholds.
What happens: risk alerts go to the CSM with a play and a deadline; expansion alerts go to the account owner with the evidence; strategic accounts also notify a leader. Unworked alerts escalate.
System role: make sure the right person hears about the account early enough, with a specific next step instead of a color.
Owner: CS managers for risk plays; sales or account management leaders for expansion.
What happens: every play and every renewal outcome is logged against the signals that preceded it, and the model is reviewed each quarter.
System role: show which signals actually predict churn and expansion in your business, and retire the ones that do not.
Owner: RevOps runs the review with CS and finance.
The board narrative
Three statements make an early-warning system legible to a board that already watches NRR.
Retention and expansion risk are now scored continuously from product usage, support sentiment, relationship changes and billing, on one account record. Every alert has an owner and a play, and every outcome is logged against the signals that preceded it.
We used to learn about churn at the renewal call and about expansion at procurement. Our goal is to see both a quarter or more ahead, turning the renewal forecast from a set of opinions into a number with a reason behind every account. We only claim that lead time when our own outcome history supports it.
We report the share of churned accounts that were flagged at least 90 days ahead, plays completed on time, expansion pipeline sourced from alerts, and gross and net retention by cohort. Flag coverage and on-time plays should rise first; retention follows.
Illustrative example, with made-up round numbers: a $20M ARR company with 101% NRR, losing $2M a year to churn and downgrades and adding $2.2M in expansion, backtests its new scores and finds that two-thirds of last year's churned dollars were visible at least 90 days ahead. Saving a quarter of that flagged churn alone retains about $333,333 and adds roughly 1.67 percentage points to NRR; pulling additional expansion forward could raise it further without a single new logo. Your figures will differ.
Cross-domain: what an early-warning system connects
The early-warning system is two VANDFORT systems working on the same account record. The Churn Signal Watchtower watches usage, ticket sentiment and relationship changes continuously and routes risk to the CSM who can act on it. Renewal Radar works the calendar: it opens each renewal early enough to run a play, with the account's risk and expansion evidence attached, so no renewal arrives as a surprise.
Both depend on the rest of the engine. Expansion alerts become pipeline that the Forecast Assistant can count, and the retention picture flows into the Board Report Engine, so the NRR slide explains itself rather than being rebuilt by hand each quarter. See all systems, or the CS Operations domain.
For who should own this kind of build as the team grows, see the GTM engineer vs. RevOps manager vs. growth engineer decision tree. Our approach is forward-deployed engineering: build inside your existing CRM, test against your own past renewals, and switch each system on only when it proves itself.
Sources: SaaS Capital, 2025 B2B SaaS Retention Benchmarks (September 2025; more than 1,000 private B2B SaaS companies surveyed in Q1 2025). Benchmarkit with Maxio, 2025 SaaS Performance Metrics Benchmark Report (May 2025; 2024 data; 583 participants), with Maxio's July 2025 summary. Gartner, survey of 243 CSOs and senior sales leaders (fielded October to November 2024; published May 2025). Hook, Customer Success in 2025 report announcement (figures confirmed in Hook's public announcement; supplied preview unavailable; sample size not established). Validity, The State of CRM Data Management in 2025 (July 2025; 602 CRM users and stakeholders). U.S. Bureau of Labor Statistics, Employee Tenure in 2026 (September 2026). The NRR Early-Warning System, the signal taxonomy, thresholds and the worked example are suggested starting points and illustrative figures, not benchmarks.
