Updated November 2026. New here? Subscribe to the SaasFlywheel newsletter for one retention teardown a week.

What Is Gross Revenue Retention? The Formula in One Paragraph

Gross revenue retention (GRR) is the share of a cohort's starting recurring revenue that is still there after a period, usually 12 months, counting only churn and contraction. The formula is (starting MRR - contraction MRR - churn MRR) / starting MRR, and by construction it cannot exceed 100%. As reported by SaaS Capital on April 24, 2026, the median GRR for bootstrapped private B2B SaaS companies with $3M-$20M ARR is 91%, and 100% at the 90th percentile (survey of more than 1,000 private B2B SaaS companies, described as essentially flat versus the prior year).

Treat that 91% as a number for that population only: SaaS Capital's page does not say how it handles downgrades, reactivation, price increases or involuntary churn. GRR, gross dollar retention and gross renewal rate are the same metric under three names, per ChartMogul (a vendor of subscription analytics), which also answers the “gross retention” search.

Illustrative arithmetic, not data: a cohort starts at $400,000 MRR. Churned customers account for $26,000 and contraction for $14,000, so GRR = (400,000 - 26,000 - 14,000) / 400,000 = 90.0%. A further $30,000 of expansion is excluded from GRR. On the same cohort, NRR is (360,000 + 30,000) / 400,000 = 97.5%. For NRR numbers by stage and ACV, see the NRR benchmarks article.

What GRR Excludes, and the Edge Cases That Break Your Number

The worked example above is tidy. Real billing data is not, and the number you report depends on a few decisions made before anyone opens a spreadsheet. GRR counts only losses. Expansion, new logos and reactivation are excluded entirely, not netted to zero (ChartMogul, vendor-stated), but every dollar of contraction counts. Most arguments about the metric are policy decisions on a few edge cases; where practice varies, teams differ, so pick a rule and document it.

The inclusion table

MovementCounts in GRRExcludedDepends on your policy
Churned customerYes
Downgrade or seat removalYes
Upsell or cross-sellYes
New logoYes
Reactivation of a churned customerYes (ChartMogul; no other page read addresses it)
Involuntary churn after a failed paymentOnce it becomes a cancellationGrace-window length
Annual price increasePick a rule and document it
Plan migrationDepends on net direction
Usage overage on metered plansDefine a committed or trailing-average baseline

Sources agree on the first four rows (ChartMogul, Stripe and others). The bottom rows are method advice, not a standard.

Edge cases that move the number

Metered and AI-credit pricing is the sharpest one. Usage decline on a usage-based plan is contraction, but a variable-revenue baseline makes GRR noisy. Define GRR on committed or trailing-average revenue and footnote it for the board (method advice). For the pricing trade-offs, see usage-based vs seat-based pricing.

Stripe (vendor-stated) lists including expansion in the numerator among the common mistakes, alongside a wrong revenue baseline, ignoring downgrades and partial churn, failing to segment, and treating GRR as a growth metric. ChartMogul (vendor-stated) warns against comparing your GRR with someone else's NRR. CFI warns against double-counting a customer who downgrades and churns in the same period.

GRR vs NRR: The Board Question Is the Spread, Not the Level

NRR tells the board whether the installed base grows. GRR tells it whether the base holds without help. The gap between them is revenue your expansion motion spends covering losses, so a strong NRR on a weak GRR is a dependency, not a result. For the same cohort NRR is always at or above GRR, because expansion only adds (ChartMogul, vendor-stated).

The 2x2 spread decision grid

No numeric thresholds here. Define “high” and “low” against your own segment row in the benchmark table below.

GRR levelHigh NRRLow NRR
High GRRHealthy base. Is expansion capacity the constraint on growth?Base holds but does not grow. Is it packaging, pricing or the expansion motion? See the expansion revenue playbook.
Low GRRThe leaky-base cell. Which losses is expansion covering, and for how long?Retention problem first, expansion second.

The low-GRR, high-NRR cell is the one to interrogate. ChartMogul (vendor-stated opinion) puts it bluntly: “A business with 70% GRR and 115% NRR has a real retention problem that a good expansion motion is currently covering for.” ChurnZero (vendor-stated) makes the same point from the other side: NRR-only focus can mask a retention problem. It adds that strong logo retention can coexist with weak GRR when lost or downgraded customers are large.

What to put in the board pack

Report GRR and NRR on the same cohort and the same window, side by side, exclusion policy footnoted. Keith Wallington, identified on ChartMogul's GRR page as an investor and chairperson, is quoted there: “If you chart gross revenue retention, net revenue retention, and growth rate on a single page you have all you need for a board meeting.” For the pack layout, see the board reporting dashboard.

Team direction follows the same split. GRR leaks belong to onboarding, support and product. The spread belongs to expansion and pricing. One headline number cannot be handed to both. ChurnZero (vendor-stated) adds that a GRR-only focus pushes the conversation upstream to Sales and Marketing.

The failure mode follows from the logic, not a statistic: when expansion slows, a thin GRR is exposed within one renewal cycle.

How to Measure GRR by Cohort and Segment

A single blended GRR hides where revenue leaves. Cut it by signup cohort first, then by plan, ARPA band, acquisition channel and customer segment, over a fixed 12-month window. ChartMogul says 12 months is common and shorter intervals must be used consistently.

Build the cohort GRR curve

Method advice, no data: take the MRR of every customer active at the start of each quarter, track only that group's MRR forward quarter by quarter with expansion stripped out, and plot one line per cohort.

ChartMogul (vendor-stated) gives the reading rule for the shape: “Segment GRR by cohort, plan, or acquisition channel to see exactly when in the customer lifecycle revenue disappears. A business that loses most of its GRR in month two likely has an onboarding problem; one that bleeds it steadily all year likely has a product-fit or pricing problem.” A step at renewal dates points at annual-contract churn (method advice, not from the page).

This answers the quarterly-wins pressure. A quarterly NRR figure resets every quarter; a cohort chart shows whether each older cohort's floor is holding or eroding, the compounding story a board can follow across four quarters.

Segment cuts that matter

Cut by ARPA band (the next section's table is your comparison frame), plan tier, channel and ICP-fit flag. Small cohorts make GRR noisy because one account moves it, so show dollar counts beside the percentage (method advice).

ChartMogul says it and most billing analytics tools compute GRR from MRR movements, and that the movement and cohort formulas are mathematically equivalent. If you build it in a warehouse, make the two reconcile.

GRR Benchmarks With Named Sources (and Why They Are Not Interchangeable)

A cohort chart shows your own trend, but a board will still ask what the line should look like. There is no single good GRR. The newest independent figure is SaaS Capital's April 2026 survey: 91% median for bootstrapped companies at $3M-$20M ARR. In the vendor data below, ARPA moves the answer more than anything else shown.

FigurePopulationSourceDateLabel
Median GRR 91%; 90th percentile 100%Bootstrapped private B2B SaaS, $3M-$20M ARR; more than 1,000 companies surveyedSaaS CapitalApril 24, 2026Independent lender survey
Median NRR 103%; 90th percentile 117.9%Same populationSaaS CapitalApril 24, 2026Independent lender survey
ARPA over $500/month: 88.1 / 80.6 / 66.2 (75th pct / median / 25th pct)ChartMogul ARPA bands; sample not statedChartMogul GRR pagePage updated September 2026; data vintage not statedVendor-stated
ARPA under $10/month: 62.4 / 46.8 / 35.9 (75th pct / median / 25th pct)ChartMogul ARPA bands; sample not statedChartMogul GRR pagePage updated September 2026; data vintage not statedVendor-stated
“Over 86%” as the top-tier GRR at any stage (about 14% of gross revenue lost a year)No sample statedChartMogulSeptember 8, 2026Vendor-stated

ChartMogul's page carries the four bands between these rows; open it for the rest. For NRR beyond the SaaS Capital companion figures, see the NRR benchmarks article.

75th pct
Median
25th pct
ARPA over $500/mo
88.1
80.6
66.2
ARPA under $10/mo
62.4
46.8
35.9
Lower GRRHigher GRR
Source: ChartMogul GRR page (vendor-stated), updated September 2026. Sample and data vintage not stated. Top and bottom ARPA bands only; GRR in %.

Rule-of-thumb bands (85%+, 90%+, 95%+ for some enterprise) circulate on Stripe's and CFI's pages with no named survey and no date. Treat them as folklore until sourced.

The comparability trap: a bootstrapped $3M-$20M median and an ARPA-banded vendor dataset measure different populations. Pick the row that matches your ARPA band and funding model, and say which in the board pack.

The GRR Leak Ledger: Which AI Churn Signals Actually Move the Number

Benchmarks tell you where you stand. They do not tell you where to dig. Split the gap between 100% and your GRR into four leaks. AI helps most where a leak has a behavioural leading signal (logo churn, contraction) and least where the cause is mechanical (billing failure) or definitional (price-change noise). AI changes how early you see a leak, not how GRR is calculated: GRR comes from billing movements, with or without AI.

The GRR Leak Ledger is this article's framework, not an industry term. Start at 100%, subtract each named leak, show expansion separately below the line. On the illustrative $400,000 cohort (arithmetic, not data): logo churn $18,000 = 4.5 points, contraction $14,000 = 3.5 points, involuntary churn $8,000 = 2.0 points. Total 10.0 points, GRR 90.0%. Price-change noise is 0, because it is not a loss. The $30,000 of expansion (7.5 points) sits below the line and lifts NRR to 97.5%.

LeakUsual ownerLeading signalAI-addressable
1. Full logo churnCS, productFalling login frequency, shrinking active-seat share, negative support sentiment, dropped featuresYes
2. ContractionCS, financeSeat utilisation under contract, usage decline on metered plans, admin removing usersPartial to yes: the behavioural part, not the realised seat removal
3. Involuntary churnBilling, financePayment failure, card expiryPartial at most: mostly rules and retries
4. Price-change and migration noiseFinance, RevOpsNot a leak: a definition choiceNo: fix with the policy above

The signals in row 1 are the commonly described behavioural ones; the AI churn prediction pillar and churn warning signs cover them. For leak 3, ChartMogul (vendor-stated) says dunning, card-update reminders and automatic retries can recover a meaningful share; the involuntary churn article carries the vendor-reported shares of total churn, which are not GRR points and should not be converted into them.

What does not move GRR: expansion propensity and upsell scoring. They raise NRR, which is where the usual “AI for retention” pitch conflates the two. Tactics live in how to reduce SaaS churn with AI.

No independent figure on GRR uplift from churn models was found, so this article states none. AI-agent products that churn on perceived value are a churn and contraction driver; see why AI agents are churning SaaS users.

Using GRR as a Forecast Floor and a Quarterly Scoreboard

Once the biggest leak is named, the same number does a second job in planning. GRR is the one retention input to a revenue forecast that does not depend on an expansion motion or an AI feature landing, which makes it the natural floor when AI tool ROI is high-variance. Use a range, not a point.

Method: next-period revenue from the existing base = starting MRR x cohort GRR, with low, base and high values taken from your own trailing cohorts. Illustrative arithmetic: a $500,000 base with an invented 86% / 90% / 93% range gives $430,000 / $450,000 / $465,000 from the existing base. That range is made up for the sum and is not a benchmark. Layer expansion and new business on top as separate scenarios. The full model is in the revenue forecasting article, and the AI ROI piece covers the variance.

For the quarterly-wins conversation, the controllable lever is contraction and early churn in the newest cohort, which a team can move without expansion luck. Show the four-quarter cohort chart as the compounding view. That is advice, not a claim about boards in general.

This is a measurement and planning aid, not financial advice. Verify the numbers against your own billing data.

Frequently Asked Questions

How do I calculate gross revenue retention?

Subtract contraction MRR and churned MRR from starting MRR, then divide by starting MRR. Expansion, new business and reactivation stay out. The worked example in the first section gives 90.0% on illustrative numbers.

What is the difference between GRR and NRR?

GRR counts only churn and contraction. NRR adds expansion, so on the same cohort it is at or above GRR (ChartMogul, vendor-stated). Stripe (vendor-stated) frames GRR as defensive and NRR as offensive.

Can gross revenue retention be over 100?

No. GRR can never exceed 100% (Stripe and ChartMogul, both vendor-stated). A figure above 100% is NRR or a data error.

Does gross revenue retention include downsell?

Yes. Downgrades count as contraction (ChartMogul and Stripe, vendor-stated), while upsell is excluded.

What is a good gross revenue retention rate?

SaaS Capital reported a 91% median for bootstrapped $3M-$20M ARR SaaS companies on April 24, 2026. Match your ARPA band and funding model against the dated table above before judging your own number.

The Bottom Line: Report GRR Beside NRR, Then Fix the Biggest Leak

Report GRR beside NRR on the same cohort and window. Cut it by cohort and ARPA band so the board sees where revenue leaves. Then fix the largest leak on the ledger first.

Keep reading in the retention hub, or start with the AI churn prediction pillar. Updated November 2026.

Sources