Who it's for: SaaS finance and FP&A team members who calculate retention metrics for the board, investors or a fundraising data room.
What you'll get: the common definitions of net revenue retention (NRR) and gross revenue retention (GRR), the variations you'll meet, a step-by-step cohort calculation, a worked MonthEndClub Co Ltd example, and a key risks and controls table.
Reading time: about 10 minutes. Retention of the content not guaranteed.
Why retention metrics get scrutinised
NRR and GRR show whether existing customers are staying and growing. Investors often weight them heavily because they say a lot about product fit and future growth. They're also easy to get wrong: include new customers by mistake, or pick a flattering method, and the number can look several points better than it is. A clear definition and a cohort-based calculation make the figure defensible.
Key terms
- Cohort: a fixed group of customers. For retention, it's usually every customer with ARR at the start of the measurement period.
- Net revenue retention (NRR, also NDR or net dollar retention): how much of the cohort's starting ARR you still have after expansion, contraction and churn.
- Gross revenue retention (GRR): how much of the cohort's starting ARR you kept, ignoring expansion. It can't go above 100%.
- Expansion, contraction, churn: ARR increases, decreases and complete losses for customers in the cohort (see the ARR bridge guide).
- Logo retention: the percentage of customers (by count, not value) who are still customers.
The formulas
NRR = (Starting ARR of cohort + Expansion − Contraction − Churn) ÷ Starting ARR of cohort
GRR = (Starting ARR of cohort − Contraction − Churn) ÷ Starting ARR of cohort
Another way to write NRR: cohort's ARR today ÷ cohort's ARR 12 months ago. Both give the same answer if the bridge is built correctly.
The most important rule: new customers who joined during the period are excluded. Their ARR is in your total closing ARR, but not in the retention calculation.
The common variations
| Choice | Common approach | Variations you'll see |
|---|---|---|
| Period | 12 months, measured at a month-end or quarter-end | Monthly or quarterly NRR, sometimes annualised |
| Method | Point-in-time cohort: customers at the start, compared 12 months later | Average of the last 12 monthly figures; or a weighted average across several cohorts |
| Base | ARR | MRR, or recognised revenue (common in public company disclosures) |
| Usage revenue | Excluded, or committed minimums only | Trailing usage included (common for usage-priced businesses, and makes NRR more volatile) |
| Customer level | Net per customer | Product-level, which can make expansion and contraction both bigger |
| GRR cap | Each customer's retained ARR capped at its starting ARR | Rarely uncapped (which turns it into NRR) |
| FX | Constant currency | Reported currency, with FX in the result |
| Small customers | All customers included | Some report retention only for customers above an ARR threshold. If so, it needs saying clearly |
Monthly and annual figures don't convert by multiplying or dividing by 12. Retention compounds. A steady monthly NRR of 99.5% gives an annual NRR of about 99.5%12 = 94.2%, not 94%.
AI shortcut: if you're asked to compare your NRR with a peer's, give an approved AI assistant the peer's published definition (from their annual report or investor materials) and yours. Ask it to list every difference in method, base, period and exclusions. It turns a misleading comparison into a footnote you can defend. Check its reading against the source text.
How to calculate NRR and GRR
| Step | What "done" looks like | |
|---|---|---|
| ☐ | Pick the measurement date and the start date 12 months earlier | Dates written at the top of the workings |
| ☐ | Export the customer-level ARR list at the start date | Total agrees to the ARR reported at that date |
| ☐ | Mark these customers as the cohort | Fixed list; nobody added later |
| ☐ | Export the customer-level ARR list at the measurement date | Total agrees to closing ARR reported for that month |
| ☐ | Match the two lists on customer ID, after mapping any ID changes or mergers | Every cohort customer has a closing value (zero if churned) |
| ☐ | Classify each cohort customer's change: expansion, contraction, churn or no change | Every customer in exactly one category |
| ☐ | Calculate NRR and GRR with the formulas above | Both figures, with workings |
| ☐ | Reconcile: cohort closing ARR + new customer ARR = total closing ARR | Difference is zero |
| ☐ | Write the definition in the pack footnote | Period, method, base and exclusions stated |
| ☐ | Get it reviewed | Reviewer sign-off on the workings |
AI shortcut: give an approved AI tool the two customer lists and ask it to flag cohort customers with no match at the closing date, and closing customers with no match at the start but a similar name. Those are usually ID changes or mergers that would otherwise show as churn plus new. A person confirms each match before the calculation is run.
Worked example: MonthEndClub Co Ltd, 12 months to 31 October 2026
MonthEndClub Co Ltd (fictional) measures 12-month, point-in-time NRR and GRR on ARR, net per customer, with all contracts in sterling.
- At 31 October 2025, it had 120 customers with total ARR of £800,000. That's the cohort.
- Over the next 12 months, cohort customers expanded by £96,000 in total.
- Cohort customers who stayed but reduced spend contracted by £24,000 in total.
- 12 cohort customers left, with starting ARR of £56,000 between them.
- New customers won during the 12 months had ARR of £192,000 at 31 October 2026.
Step 1: the cohort bridge
| Cohort at 31 October 2025 | £ |
|---|---|
| Starting ARR of cohort | 800,000 |
| Expansion | 96,000 |
| Contraction | (24,000) |
| Churn | (56,000) |
| Cohort ARR at 31 October 2026 | 816,000 |
| Add: new customers (excluded from retention) | 192,000 |
| Total ARR at 31 October 2026 | 1,008,000 |
The total agrees to the closing ARR in the October pack, so the cohort and the new customers together account for every pound.
Step 2: the metrics
| Metric | Workings | Result |
|---|---|---|
| NRR | (800,000 + 96,000 − 24,000 − 56,000) ÷ 800,000 = 816,000 ÷ 800,000 | 102.0% |
| GRR | (800,000 − 24,000 − 56,000) ÷ 800,000 = 720,000 ÷ 800,000 | 90.0% |
| Logo retention | (120 − 12) ÷ 120 = 108 ÷ 120 | 90.0% |
| Total ARR growth (for contrast) | 1,008,000 ÷ 800,000 − 1 | 26.0% |
Step 3: the classic mistake
If someone divides total closing ARR by total opening ARR and calls it NRR, they get 1,008,000 ÷ 800,000 = 126%. That's growth, not retention. It overstates NRR by 24 percentage points because it counts £192,000 of new customers as if existing customers had bought more.
Step 4: why the GRR cap matters
Suppose one cohort customer dropped a £10,000 product but added a £15,000 one. Net per customer, that's £5,000 of expansion and no contraction, so its retained ARR for GRR is capped at its starting value. If the calculation was done at product level instead, it would show £10,000 contraction and £15,000 expansion: NRR is unchanged, but GRR falls. Neither is wrong, which is why the method belongs in the footnote.
No journals required
NRR and GRR are KPIs worked out from customer data. They don't create entries in the ledger. The link to the accounts is the reconciliation of total closing ARR to MRR and then to revenue, covered in the ARR vs MRR vs revenue guide.
If you report under US GAAP (ASC 606)
Retention metrics aren't defined by IFRS 15, FRS 102 or ASC 606. US-listed companies that disclose NRR typically explain how they calculate it, and SEC staff guidance on key performance indicators expects a clear definition and consistent use over time. Many US-listed SaaS companies use a trailing or revenue-based method, so their figures may not be directly comparable with a point-in-time ARR method.
Common mistakes and audit watch-outs
- Including new customers. The biggest and most common error. The cohort is fixed at the start date.
- Annualising monthly retention by multiplying. Retention compounds; use the power, not ×12.
- Changing method when the number dips. Investors notice. If the method changes, restate prior periods and say so.
- Customer ID changes counted as churn plus new. This understates retention. Map IDs before calculating.
- Excluding small customers without saying so. Fine if disclosed; misleading if not.
- Trusting an AI-calculated NRR without the reconciliation. AI tools can quietly include new customers or mis-join lists. The check that cohort closing ARR plus new ARR equals total closing ARR catches most of this.
Tools that can help
Billing platforms such as Chargebee and Stripe Billing report retention, but check their method (cohort, base, usage treatment) against your written policy before quoting them. Many finance teams calculate NRR in Excel or Google Sheets from customer-level exports, which keeps the cohort visible. Reporting tools such as Fathom help present the trend. A tool's number still needs the reconciliation and footnote.
Key risks and controls
| Risk | Control | How to evidence it |
|---|---|---|
| New customers included in NRR | Fixed cohort list from the start date | Cohort export saved with date and total agreed to reported ARR |
| Calculation doesn't tie to reported ARR | Cohort closing + new = total closing ARR check | Reconciliation line in the workings, reviewed |
| ID changes distort churn | Customer ID mapping maintained by finance | Mapping table with dates and approver |
| Method changes hide a decline | Written definition; change log; restated comparatives | Footnote in the pack and dated change log |
| AI-produced figures not checked | AI used for matching and flagging only; formulas in a reviewed sheet | Reviewer sign-off on the final workings |
Key takeaways
- NRR = (Start + Expansion − Contraction − Churn) ÷ Start. GRR = (Start − Contraction − Churn) ÷ Start.
- Fix the cohort at the start date. New customers are never part of retention.
- Write down the period, method, base, usage treatment and any exclusions, and show them in a footnote.
- Reconcile cohort closing ARR plus new customer ARR back to total closing ARR every time.
- Monthly retention compounds; don't multiply it by 12.
FAQs
What's a good NRR for SaaS?
Benchmarks vary widely by customer size and pricing model. Above 100% means existing customers are growing in total; many enterprise-focused SaaS businesses report well above that, while small-business products often report below. Comparisons only work if the methods match.
Can GRR be higher than NRR?
No. GRR ignores expansion, so it's always equal to or lower than NRR. If your GRR is higher, something in the classification has gone wrong.
Should we use ARR or recognised revenue?
Private SaaS companies most commonly use ARR. Some public companies use recognised revenue over a trailing period. Either works if it's stated and applied consistently.
Can AI calculate our retention?
An approved AI tool can do the matching, flag likely ID changes and draft the calculation. The cohort, method and final number are checked by a person against a formula-based sheet, including the reconciliation to total ARR.
Related guides
- How to Build an ARR Bridge (New, Expansion, Contraction, Churn)
- ARR vs MRR vs Revenue: Why They Don't Match and How to Reconcile Them
- SaaS Metrics Formula Sheet (CAC Payback, LTV, Magic Number, Rule of 40)
- Monthly Management Accounts and Board Pack Layout
More in Metrics and Reporting guides.
From the Club
100% retention rate among people who own one. Probably. The Washed Cap · The Dad Hat
Join the Club. New guides, early access and members-only batches. Join here.
This guide is general information about common practice. It isn't tailored to any business, and standards, tax rules and figures change. Check your own circumstances, the latest rules and, where needed, a qualified professional before acting.
Last reviewed: October 2026.