Key Takeaways
Slide 14 of the board deck is retention. It shows net revenue retention at 104%, in green, with a two-year trend line pointing up. It doesn’t show gross revenue retention. Slide 22, the finance appendix nobody reaches, lists 11 logos that didn’t renew this quarter and 6 that renewed smaller. Both slides are accurate. Only one of them describes what customers decided.
That gap between the number on the retention slide and the number in the appendix is the subject here. Net revenue retention is the metric the company chose to publish. Gross revenue retention is the one it mostly didn’t, and the one that moved in 2025. For the leader who owns support, onboarding and self-service, that’s not a finance curiosity. The gross line is the only retention number her function can move, and it’s the one nobody is asking her about.
Gross revenue retention is the number companies choose not to publish
Ordway reviewed the filings of 135 SaaS, cloud and fintech companies listed on the NASDAQ and NYSE. Ninety reported net retention only. Seven reported both net and gross. Three reported gross only. Thirty-five reported neither. Gross retention, the metric that counts only what customers kept, appears in about 7% of public disclosures. And the net figure is thinning: 27 public SaaS companies reported net dollar retention in the first quarter of 2026, down from 32 the quarter before.
The exception is instructive. Procore’s 2025 annual report states: “We consider gross retention rate to be a key metric… Our GRR reflects only customer losses and does not reflect customer expansion or contraction.” It reports GRR of 95%, 94% and 95% across three years, and adds that it doesn’t consider NRR a key metric “due to the impact of pooled volume contracts”. A construction-software company decided the net number told investors less than the gross one, and said so in a filing.
HubSpot shows the size of the gap at a company that reports both in different forms. On its Q4 2025 call the CFO put customer dollar retention “in the high 80s” and net revenue retention at 105%. Roughly 15 points between what customers kept and what the company reported as retention. Neither number is wrong. They answer different questions.
Net revenue retention measures your contract structure at least as much as your customers
Procore’s reason for demoting NRR deserves a closer look, because the same mechanism runs through every NRR figure you’ll be shown. A multi-year pooled volume contract produces 100% NRR whatever the customer does with the product. Snowflake’s 126% is calculated as a two-year cohort ratio of product revenue in which churned customers stay in the calculation at zero, a definition no ARR-based survey benchmark shares. Workday’s 98% gross retention excludes volume and price adjustments. Benchmarkit’s 2026 data splits the market by pricing model: usage-priced companies at 108% NRR, seat-priced at 98%. Ten points of “retention” decided by billing design, before any customer forms an opinion.
Then there’s what gets left out. Ordway found about half of the 97 companies that track retention use a non-standard formula or exclude some customers. Eight formulas. Six definitions of recurring revenue. 29% smooth over trailing periods. More than 30 names for the metric. The most commonly excluded groups are small businesses and customers acquired through self-service channels. The customers most likely to churn are the ones most likely to be outside the number.
Investors have noticed. Ray Rike’s analysis of Meritech’s regressions shows NRR’s explanatory power for public cloud multiples falling from an R² of 0.40 in July 2021 to 0.07 in January 2023, recovering to 0.18 by May 2024. Revenue growth sat at 0.33 and Rule of 40 at 0.28. The band premium still exists: SEG’s index shows companies above 120% NRR trading at a median 9.3x revenue against 3.1x below 100%. But “NRR drives the multiple” is a 2021 sentence. If you’re building an internal case for investing in the gross line, this is the argument a CFO recognises. The net number has become a weaker signal. The growth story NRR tells rests on a gross foundation that has to hold first.
What gross revenue retention measures, and where the definitions disagree
Gross revenue retention starts with recurring revenue at the beginning of a period, subtracts what was lost, and divides by the starting figure. Expansion never enters. The ceiling is 100%. It answers one question: of what customers were paying you a year ago, how much are they still paying?
The disagreement is over “lost”. SaaS Capital calculates gross retention by capping each customer’s current revenue at its prior-year level, so a downgrade counts as loss; Ordway and Software Equity Group count contraction the same way. Procore’s filing counts only customers who left. Two companies with identical logo churn and different seat-reduction rates will report the same GRR under one definition and numbers several points apart under the other. The seat reductions your renewals team fights hardest over are the exact line the definitions disagree about. Before comparing your figure to any benchmark, ask which one it uses. Most don’t say.
That’s also why published medians range so widely. They aren’t the same measurement.
Same three letters, five different populations, five different cohort methods. Pick the row whose sample looks like your company, and hold your own definition constant from quarter to quarter.
The 2025 decline in gross revenue retention, by segment
Benchmarkit’s longitudinal series is the one to watch because it measures the same way each year. GRR drifted from 90% to 88% over three years, then dropped to 84% for calendar 2025, with the 75th percentile falling from 95% to 91%. When the top quartile loses four points along with the median, it isn’t a story about weak operators. It’s the market.
The segment cuts show where. Aleph’s breakdown of the same data puts GRR at 91% for $50–100K contracts and 80% below $5K; 88% for sales-led companies, 80% hybrid, 79% product-led; 89% for companies growing 11–20% and 79% below 10%. SaaS Capital’s 1,500-company survey found the same ACV gradient two years earlier, 90% below $12K rising to 93% above $250K, and called 90% “table stakes” for performance parity. Its lending criteria require 85%. Below that line, retention stops being a metric and becomes a covenant.
Two mechanisms explain most of the drop, and both land in customer operations. The first is contraction. ChartMogul’s billing data shows churn is the largest source of lost ARR at every price point, but at higher price points contraction can reach 40% of all ARR lost. Buyers are actively reclaiming seats: Zylo, tracking 40 million licences, reports utilisation rising from 47% in 2024 to 54% in 2025 as companies cut what nobody uses. Every unused seat a customer finds is a downgrade on your side. The second is the long tail of small, self-serve, product-led accounts, where GRR sits near 79–80% and where churn is higher than it should be for reasons that trace to the first weeks, not the renewal call.
Why gross revenue retention belongs to customer operations
Expansion is sold. Retention is delivered. A customer who renews flat did so because the product kept working, the questions got answered, and the people using it would have noticed if it left. Every one of those is a service outcome, which makes the gross line the customer-operations number and the net line a sales number wearing a retention label.
The evidence for what moves it is thinner than the volume of advice suggests, but what exists points the same way. The one randomised experiment in the literature, at a public cloud provider, gave 366 of 2,673 new customers proactive guidance on basic features in their first days. They churned at half the rate in week one, asked 19.55% fewer support questions, and used 46.57% more of the service over eight months. Retention and support cost moved together, not against each other, and the window was the first week. Gartner’s survey of 5,728 customers shows the same lever from the failure side: 43% of customers who tried self-service couldn’t find content relevant to their issue, and only 14% of issues resolved there. A new account that can’t find the answer to its first question is a gross-churn event twelve months early.
ChurnZero’s 793-company data adds correlational support with a caveat worth keeping: NRR was 98% where a support function existed against 93% without, and 99% against 94% with a customer enablement function. Companies that fund those roles are also bigger and healthier, so read it as direction. The same study found dedicated renewals teams correlate with lower NRR, which the authors attribute to companies formalising renewals once risk is already high. Adding a renewals desk treats the symptom. The cause sits upstream, in the lifecycle stages the service function runs.
So decompose the gross line by cause, and route each cause to the owner who can act on it.
| Gross loss | What it usually means | Owner | Leading signal |
|---|
| Logo churn in year one | Never reached first value | Onboarding | Milestone slippage; first-week questions unanswered |
| Logo churn after year one | Dependency never formed, or a champion left | CS and support | Days since last positive interaction; users stop searching |
| Contraction | Seats or modules nobody used | Value realisation, pricing | Utilisation by account; feature questions that never come |
| Silent churn | Couldn’t find the answer, never asked | Knowledge and self-service | Failed searches; repeat questions; zero tickets |
The fourth row is the one no team owns by default and the one Gartner’s 43% describes. It only becomes visible when failed searches, tickets and onboarding status are readable against the same account. A health score built on that record catches it months before renewal; one built on logins catches it at the exit interview.
Report gross revenue retention before finance does
You don’t need a new definition from finance to start. Three numbers, every quarter, on one page, next to the NRR slide. First, gross revenue retention with contraction included, which is the SaaS Capital and Ordway convention and the harsher of the two. Second, the same figure excluding contraction, so you can compare to a Procore-style disclosure without arguing about it. Third, the four-row decomposition above, with an owner’s name in each row. The third number is the one that changes meetings, because it turns “improve GRR” into four assignments.
The decomposition needs the churned account’s history in one place. That is where most attempts stall: searches in the help-centre tool, tickets in the help desk, onboarding in a project tracker, usage in an analytics product. In MatrixFlows the account, its requests, its onboarding milestones and its knowledge activity are one linked record. A churned logo reads as a story rather than four exports: stalled at milestone three, searched for the firmware article nine times, opened two tickets, went quiet in month seven. Search-gap analytics show which accounts are asking questions nobody has answered, which is the silent-churn row made visible. The renewal system then works from what the customer did rather than from what the CSM remembers.
Start this quarter by sorting last quarter’s lost and downgraded accounts into the four rows by hand. The pattern will tell you which owner moves first. NRR is what the company sold. GRR is what the customer decided.
Create a Free Workspace → and build the four-row decomposition on records that already connect the search, the ticket and the milestone.
.svg)
.svg)
.svg)
