How do you actually diagnose B2B SaaS churn — and what's the framework that works?
Direct Answer
Diagnose B2B SaaS churn by coding every lost account into one of five types — onboarding failure, value erosion, competitive switch, pricing rebellion, or genuine bad fit — then run a four-step diagnostic that cohorts by tenure, codes each loss, looks for clustering above forty percent, and asks a counter-factual.
Blended churn rates hide everything. A DevOps SaaS at twenty-five million ARR ran this against sixty churns and discovered forty-two percent were onboarding failures, then moved GRR from eighty-one to ninety-one percent in four quarters.
TL;DR
- Five churn types map to distinct tenure windows — diagnosis starts with when, not why.
- The four-step diagnostic forces a coded reason for every churn so patterns surface instead of anecdotes.
- Onboarding failures are roughly thirty percent of total churn industry-wide and the single most fixable bucket.
- Three failure modes hide the real cause — blended reporting, treating churn as a CS-only problem, and exit interviews with departing champions.
- Real benchmarks — SMB fifteen to twenty-five percent gross churn, mid-market eight to fifteen, enterprise four to ten.
The 5 Churn Types and Their Fixes
The single biggest unlock in churn analysis is refusing to let any account die without a type code attached. "Other" is forbidden. Below is the taxonomy, the tenure window each type clusters in, the telltale signal in the data, and the function that owns the fix.
| Type | Tenure Pattern | Signal | Owns the Fix |
|---|---|---|---|
| Onboarding failure | Months 0 to 6 | Low product adoption, no key-event in first ninety days, missed kickoff milestones | Onboarding plus Product |
| Value erosion | Months 6 to 18 | Champion role change in LinkedIn, drop in weekly active users, support tickets shift to billing | CS plus Marketing |
| Competitive switch | Months 12 to 24 | Sudden re-evaluation, competitor name in support tickets, RFP mention at renewal | Product plus Sales |
| Pricing rebellion | At renewal | Price-to-value comments in QBRs, finance buyer joins late calls, requests to downgrade tier | Finance plus CS |
| Genuine bad fit | Any time | Stage size, vertical, or use case outside ICP from day one | Sales plus Marketing |
The onboarding bucket is almost always the largest and the most embarrassing — Gainsight's 2024 CS Index puts it at roughly thirty percent of total churn, and Bessemer's State of the Cloud lines up with that range. Competitive switches sit near fifteen percent. Pricing rebellions are smaller but concentrate violently inside renewal months, which is why a moving-window dashboard misses them.
Bad-fit churn requires admitting Sales closed something they shouldn't have.
The 4-Step Diagnostic
Step one. Cohort the churned accounts by tenure. Pull every logo lost in the last four quarters and plot them by month-of-life at churn. You are looking for visual clusters.
A spike at months three to five means onboarding. A flat distribution with a spike at renewal months means pricing. A creeping rise around month eighteen means value erosion.
This single chart kills more bad theories than any other artifact in RevOps.
Step two. Code each churn by the five types — mandatory, no "other". Force the CSM, AE, and renewal owner to pick one. If they cannot, the analyst picks based on signal data. Disallowing "other" is the rule that makes everything else work — Lincoln Murphy has hammered this point for a decade.
Step three. Run the pattern. If more than forty percent of churn lands in one bucket, you have a systemic problem, not a customer problem. Forty-two percent onboarding failure does not mean forty-two unlucky customers — it means your onboarding motion is broken and every new logo is at risk.
Step four. Counter-factual. For each churn, ask the question that forces action — "Would changing X have saved this account?" If yes, X becomes a project. If no, you have learned the ceiling of what CS can do alone.
The 3 Analysis Failure Modes That Hide the Real Cause
First, reporting blended churn rate without segment cohort. A nine-percent blended number can hide twenty-two-percent SMB bleeding under five-percent enterprise stability — or the opposite. Segment dashboards by ICP tier before anyone looks at them, because the executive team anchors on the first number they see.
Second, treating churn as a CS problem when roughly thirty percent is an acquisition problem. CS cannot fix a customer who never should have been sold. If bad-fit is above ten percent, that is a Marketing-MQL and Sales-qualification conversation, not a CS accountability conversation.
Nick Mehta's team has made this argument since the original Customer Success book and it remains under-implemented.
Third, "exit interviews" with the departing champion instead of with the economic buyer. Champions tell you what you want to hear — they liked the product, they fought for it, budget got cut. The economic buyer tells you the truth — ROI never materialized, or a competitor showed up with a better case.
Insist on the buyer interview. ChurnZero's 2024 Churn Report quantifies this gap.
A real example anchors the framework. A twenty-five-million-ARR DevOps SaaS coded sixty churns across four quarters this way. Forty-two percent were onboarding failures, eighteen percent competitive switches, twenty-two percent bad fit, the remainder split between erosion and pricing.
The response was not a CS reorg — it was an onboarding redesign that cut time-to-value from ninety days to thirty-five, plus tightened Sales qualification. GRR moved from eighty-one to ninety-one percent over four quarters. Mosaic.tech-style modeling translates that ten-point move into roughly two-and-a-half million in protected ARR per year.
Tooling-wise, Gainsight's Churn Risk Score, ChurnZero's Reasons, and Catalyst's Health framework all support type-coding natively. Looker or Sigma cohort dashboards handle tenure visualization. The tool is never the bottleneck — forbidding "other" is.
Frequently Asked Questions
GRR versus NRR for diagnosing churn? Use gross revenue retention to diagnose — it strips expansion noise and shows the true bleed. NRR is a board-reporting number that can mask serious churn underneath aggressive upsell.
How long should you wait before declaring a customer churned? Ninety days past contract end with no renewal signature and no active negotiation. Anything shorter generates false positives. Anything longer corrupts the cohort math.
Are exit interviews worth doing? Yes — but only with the economic buyer and only with a structured five-type prompt. Open-ended champion interviews produce flattering noise that confirms whatever theory you walked in with.
Sources
- Gainsight 2024 Customer Success Index — churn type benchmarks across SaaS segments.
- ChurnZero 2024 Churn Report — exit interview gap and signal accuracy data.
- Bessemer Venture Partners State of the Cloud 2024 — gross churn benchmarks by segment.
- Nick Mehta, Dan Steinman, Lincoln Murphy — Customer Success, Wiley.
- Lincoln Murphy, Sixteen Ventures blog — no-"other" coding discipline and ideal customer profile.
- Pavilion 2024 RevOps Benchmark Report — segmented retention by ICP tier.
- Mosaic.tech research — financial impact modeling of one-point GRR moves.
- OpenView SaaS Benchmarks 2024 — SMB versus enterprise churn distributions.