Net Revenue Retention (NRR)

One-liner

The share of recurring revenue retained from an existing-customer cohort over twelve months, including expansion and contraction but excluding new-logo revenue.

Why it diverges

NRR is the highest-divergence metric. The same customer book reports a different retention rate on each axis: a recurring-revenue (ARR or MRR) basis is one legitimate convention among several, alongside a GAAP-revenue basis, an ACV or annualized basis, and varying timing and cohort choices. An investor recomputes against the applicable convention, with the codified recurring-revenue cohort method as the reference’s default. The new-logo exclusion is the one near-settled point.

Legitimate convention families

FamilyAxisMethod (short)Representative sourceDirection vs defensible
ARR-basis 12-month cohortbasis + timingcurrent ARR of the 12-month-prior cohort over that cohort’s ARR 12 months priorrecurring-revenue cohort methodthe defensible default
4-quarter / monthly averagetimingaverage of the monthly annualized rate over the periodmonthly-average timingsmooths spikes; can read lower or higher
trailing-revenue basisbasisGAAP revenue of the same customers, trailing-twelve vs priorGAAP-revenue basisrevenue lag can understate a fast-expanding book
calendar-revenue, excludes acquiredbasis + compositioncalendar-year revenue of the prior-year cohort, acquired customers strippedcalendar-revenue basisstripping acquired customers usually lowers it
implied-monthly basisbasisimplied monthly subscription revenue end vs one year priorimplied-monthly basissimilar to MRR; depends on revenue smoothing
segment-restricted cohortcompositiononly a healthy segment (for example “premium”) in the cohortsegment-restricted cohortrestricting to a healthy segment usually raises it

What the investor likely recomputes on

The codified convention is the basis a founder’s number gets measured against when no specific investor convention is on the table. It is the reference’s default anchor.

The codified NRR is an ARR-basis twelve-month cohort (the standards-board cohort method). Legitimate variants average it across the period or threshold it to a cohort segment, and the specific basis, timing, and cohort sophistication vary by convention, so the direction is settled but the exact read is not. Name the convention against the codified one rather than assume it.

The defensible read

“Defensible” is relative to the investor. It is the convention the raise gets judged against, which is not always the reference’s default. So the read has two tiers.

Under either anchor, the aggressive convention is whichever inflates via composition (segment-restricting the cohort) or a flattering timing choice; that is always the labeled risk, not the defensible read. Principle: defensibility, not optimization, the number that survives an investor’s recompute. A defect classification is absolute and does not move with the convention anchored on.

Aggressive-vs-defect test

The predicate for classifying an NRR spread:

Canonical formula

NRR=BOP value+ExpansionContractionChurnBOP value(existing customers only; “value” = ARR, revenue, or implied-MRR depending on the basis)\text{NRR}=\frac{\text{BOP value}+\text{Expansion}-\text{Contraction}-\text{Churn}}{\text{BOP value}}\quad(\text{existing customers only; ``value'' = ARR, revenue, or implied-MRR depending on the basis})

Primary source: the Standards Board codified cohort method, “Net Revenue Retention (NRR), also known as Net Dollar Retention (NDR), measures the percentage of recurring revenue retained over a specific period,” recorded in the reference’s verified evidence set, sourced from the SaaS Metrics Standards Board’s net revenue retention page. The codified method is the authority here.

Standards-board cohort method (the defensible default): current value of the cohort of all customers as of 12 months prior to the period end, over that same cohort’s value 12 months prior. New logos are excluded by construction.

InputSource-system classTiming-offset
BOP value (the cohort denominator)billing / data warehouse (recurring-revenue book)as of 12 months prior to the period end
Cohort membership (existing customers only)CRM / billingfixed at 12 months prior; held constant across the window
Expansionbillingover the trailing 12-month window
Contractionbillingover the trailing 12-month window
Churnbillingover the trailing 12-month window
Numerator (current value of the same cohort)billing / data warehouseas of period end

Reconciliation note

Composition-closed, basis/timing-bounded. The composition axis closes to a zero residual: from the reported aggregates we can confirm the cohort is existing-customers-only with new logos out and expansion and contraction handled, and that closes from the aggregates. Basis and timing do not close. They interact non-additively (an ARR basis read point-in-time and a GAAP-revenue basis read as a trailing-twelve average are not a sum of separable adjustments), so they are reconciled bounded and qualitative: name the convention the founder used against the defensible ARR-basis cohort, and bound the direction and magnitude of the spread. The per-axis reconstruction of that spread from the client’s raw source systems is out of scope for this reference. Reconstructing it from raw source systems is a separate, multi-week effort the reference does not perform.

The three axes

NRR diverges on all three axes, the only metric in the library to do so materially. Composition is the settled part: existing-customer cohort only, expansion and contraction in, new logos out, with the new-logo exclusion consistent across conventions. Basis splits widest, recurring (ARR or MRR) vs GAAP-revenue (trailing-twelve-month or calendar) vs ACV or contracted vs implied-monthly. Timing is reported so inconsistently that two independent readings often cannot classify it the same way.

See the three axes for the shared model.

Classic errors touching NRR

Scope boundary

IN SCOPE (from the ~12 reported aggregates), the recognition side: the composition axis, confirm new logos are excluded and expansion and contraction are handled. OUT OF SCOPE (named but not computed): basis and timing, because they require raw customer-level data. The per-axis reconstruction from the client’s raw source systems is out of scope for this reference. Reconstructing it from raw source systems is a separate, multi-week effort the reference does not perform.

Citations

Cite this definition

Sculpted's SaaS Metrics Reference · Net Revenue Retention · v1.0.0
sha256:d56a44eb71027d6448851060b40e43954f3f98ef2dcc3c2d745de0694864e883
https://sculpted.io/reference/nrr

The hash changes only when the definition changes.

copied