Quick definition: Net revenue retention (NRR) measures how recurring revenue from a starting customer cohort changes after expansion, contraction, and churn, excluding revenue from newly acquired customers.
What is NRR?
NRR shows whether an existing revenue base grows or shrinks over a stated period. It follows the customers present at the beginning, then includes their upgrades, downgrades, and cancellations. By excluding new customers, it isolates retained customer economics from acquisition. A result above 100% means expansion and reactivation exceeded lost and contracted recurring revenue within the cohort.
NRR differs from gross revenue retention (GRR), logo retention, renewal rate, churn rate, ARR, and MRR. GRR excludes expansion, so it focuses on preserving starting value. Logo retention counts customers rather than dollars. Renewal rate is usually the share renewing at a contract event. MRR and ARR are levels of recurring value, while NRR is a cohort change ratio. Definitions vary, so labels must state period and treatment of reactivation, usage revenue, and foreign exchange.
NRR formula and cohort
NRR = (starting recurring revenue + expansion + reactivation − contraction − churn) / starting recurring revenue × 100. Include only customers in the starting cohort. A new account’s revenue is excluded even if it looks similar; an existing account’s upgrade is included. Select a fixed starting date and a consistent ending window, commonly monthly, quarterly, or trailing twelve months.
For a meaningful denominator, use starting recurring revenue from the cohort, not ending revenue or all current customers. State whether amounts are gross or net of credits, whether a pause is churn, and how acquisitions, account merges, and contract amendments are treated. Finance and product should agree on the cohort ledger.
NRR measurement policy
A reliable NRR series starts with an immutable cohort snapshot. On the selected start date, store each account identifier, parent-account relationship, plan, recurring value, currency, and status. Do not add accounts that purchase later, even if the sales team attributes them to an existing customer. When accounts merge or split, preserve a documented mapping so the same economic customer is neither double-counted nor accidentally treated as new revenue.
Classify each change against the starting cohort. A higher commitment from an existing account is expansion; a lower commitment is contraction; a fully lost commitment is churn; a return after a documented lapse is reactivation. Decide whether a temporary payment failure, suspension, or downgrade to a free tier is churn at the event date or only after a grace period. For annual contracts, specify whether the bridge is based on normalized MRR, ARR, or contract value, and do not combine the measures without a label.
Publish a bridge alongside the percentage: starting cohort revenue, expansion, contraction, churn, reactivation, ending cohort revenue, and NRR. That makes a 102% result interpretable. It could represent excellent broad expansion, or a large loss offset by one major account upgrade. A constant-currency bridge can separate customer behavior from exchange-rate movement; a reported-currency bridge remains important for financial planning. Both require the same policy over time.
NRR in A/B testing
NRR is a valuable north-star or long-term metric for changes to onboarding, product value, support, pricing, expansion, and renewal flows. It is often unsuitable as the only short-horizon primary metric because renewals mature slowly and revenue is concentrated. Use proximal outcomes such as activation, successful workflow completion, or renewal intent with a documented causal pathway, then maintain a holdout or follow cohorts into NRR.
A renewal offer test should estimate revenue per assigned eligible account, including accounts that do not respond, rather than compare only purchasers. Track cancellation, discount cost, support contacts, and complaints as guardrails. Preplan enough duration to observe the relevant cycle; this test-duration guide explains the maturity problem, and confidence intervals prevent overreading a small cohort.
Worked scenario
A SaaS company begins a quarter with a cohort worth $1,000,000 MRR. Within that cohort, $90,000 expands, $20,000 reactivates, $40,000 contracts, and $100,000 churns.
NRR = (1,000,000 + 90,000 + 20,000 − 40,000 − 100,000) / 1,000,000 × 100 = 97%
The company acquired $150,000 of new MRR during the quarter, but it does not enter NRR. A new onboarding experiment may help future NRR, yet this quarter’s 97% result cannot establish causation without a counterfactual. The team follows randomized cohorts and evaluates whether early activation lifts lead to later retained revenue.
Suppose instead that one account worth $250,000 expands by $80,000 while twenty smaller accounts worth $5,000 each churn. The revenue bridge can still show a healthy NRR even though logo retention and product breadth are worsening. Management should inspect GRR, customer count retention, concentration, and the distribution of expansion in parallel. NRR is a valuable summary, but it cannot tell whether durable value is broad-based or dependent on a small number of contracts.
Data-quality limitations
NRR depends on accurate customer identity and contract history. Account merges, parent-child billing, delayed cancellation dates, credits, multi-currency conversions, and backdated amendments can change both cohort and revenue bridge. Use a reconciled subscription ledger, preserve historical cohort membership, and version policy changes. Client-side activity events are not a replacement for billing records.
For experiment follow-up, join assignment to the account ledger before treatment and check for differential attrition or exposure. An allocation fault can bias the cohort; investigate unusual split counts with SRM diagnostics. Interpret segments cautiously because a few large accounts can dominate the rate.
Timing creates another limitation. A cohort measured before contract renewal has not yet had the opportunity to churn, while a cohort measured after renewal may include late amendments and credits. Compare cohorts at equivalent lifecycle age and retain a clear “as of” date. Do not backfill an NRR claim using account states that were unavailable at the decision date without identifying the revision. Sales compensation changes, packaging migrations, and account-management interventions can also alter the bridge independently of a product experiment.
Common mistakes
- Including new-customer revenue: that turns NRR into a growth measure.
- Confusing NRR with GRR: GRR excludes expansion.
- Changing cohort rules mid-series: comparisons need stable definitions.
- Ignoring large-account concentration: inspect the revenue bridge and distribution.
- Using it as instant experiment proof: it needs time and a causal comparison.
Frequently asked questions
Can NRR exceed 100%?
Yes. Expansion and reactivation from the starting cohort can exceed contraction and churn.
Does NRR include new sales?
No. New-customer revenue is intentionally excluded.
What is the difference between NRR and GRR?
NRR includes expansion; GRR measures retained starting revenue before expansion.
Why can NRR improve while customer retention falls?
A few expanding high-value accounts can outweigh churned lower-value accounts.
Should reactivated customers be included?
They can be included as reactivation if the policy treats them as members of the original cohort. Report the component separately so readers can see its contribution.
How often should NRR be calculated?
Monthly monitoring is common, but quarterly or trailing-twelve-month views can reduce contract-timing noise. Use the cadence that fits the renewal cycle.
Can an experiment use NRR as its primary metric?
Only when traffic, account count, and follow-up time can support a precise causal estimate. Most product tests need a preplanned proximal primary metric and NRR follow-up.
Summary
NRR measures how a fixed starting revenue cohort changes after expansion, contraction, churn, and reactivation. Its reliability depends on a stable cohort ledger and transparent revenue policies. In product experiments, use it as mature evidence of durable value rather than a shortcut for short-term causality.