Net revenue retention explained with a worked example

Net revenue retention, or NRR, measures how revenue from a defined group of existing customers changes after expansion, contraction and customer losses. It is a revenue measure, not the percentage of customers who stayed and not a forecast of total-company growth.

In plain language, it asks: before adding new customers, is the established customer base spending more or less?

Define the customer group before calculating the ratio

Choose a base-period customer group and measure that same group’s revenue in the comparison period. Customers who leave remain in the calculation with the revenue they actually contribute, which can be zero. Dropping them would hide churn.

The simplest form is:

NRR = current-period revenue from the original customer group ÷ base-period revenue from that group × 100.

This generic example assumes a fixed perimeter, positive base revenue and comparable periods. Actual issuer definitions can use different eligibility rules, averaging methods, currencies and revenue categories. Record those choices beside the percentage.

The arithmetic behind 120 percent NRR

Every figure in the following table is hypothetical, not an observation about Anthropic or another company. The same customer group generated ₹100 lakh in the earlier period.

Movement within the original group	Revenue effect
Base-period revenue	₹100 lakh
Additional spending by existing customers	+₹30 lakh
Reduced spending by existing customers	−₹5 lakh
Revenue lost from customers who leave	−₹5 lakh
Current-period revenue from the original group	₹120 lakh

NRR is ₹120 lakh divided by ₹100 lakh, multiplied by 100: 120%. The original group spends 20% more overall despite losses within it.

Under a conventional gross revenue retention calculation, remove the ₹30 lakh of expansion. The remaining ₹90 lakh divided by ₹100 lakh gives 90% gross retention. Both readings can be correct: expansion more than offsets a leak in the base.

If new customers contribute a further ₹40 lakh, total revenue in this simplified business becomes ₹160 lakh. Its total revenue growth is 60%, not 20%. That reconciliation only works because the example explicitly assumes no other businesses, scope changes or adjustments.

Customer retention answers a different question

Suppose a business starts with ten customers and loses one. Its customer-count retention is 90%. Its NRR could still be high if the remaining customers expand substantially.

The reverse is also possible. A business can retain every customer while each customer spends less, producing strong customer-count retention but revenue contraction.

Measure	What is retained	Includes expansion spending
Customer retention	Customer relationships, counted under a stated rule	Not a spending measure
Gross revenue retention	Revenue from the original group, excluding expansion	No, conventionally
Net revenue retention	Revenue from the original group after all movements	Yes

Do not infer customer satisfaction, profitability or competitive advantage from one of these metrics alone. Concentration can also matter: expansion at one large customer can outweigh weakness across many smaller ones.

Why the measurement window matters

A quarter, a trailing year and an average of cohort readings are different constructions. Two issuers using the NRR label may not be measuring the same thing.

For a current definition example, Anthropic’s 26 October 2026 S-1 glossary compares three-month periods a year apart and excludes specified fixed-price consumer plans from its qualified customer group. That is a perimeter choice, not a universal NRR convention. The IPO reading guide explains the wider filing context.

Other issuers’ methods must be read independently. Comparing retention percentages without matching periods and eligibility is like comparing store growth after letting each retailer select a different age threshold. Our same-store sales guide discusses that related cohort problem.

A high percentage is not a repeatable growth rate

An early pilot account can expand dramatically from a small starting bill. Once the deployment is much larger, repeating the same percentage increase requires a larger absolute spending increase. The ratio itself does not tell you whether that demand exists.

Usage-based businesses add another question: are customers consuming more because deployment is broadening, or because of a temporary project? Pricing and product mix can also change the bill without an equivalent change in underlying usage.

Ask for the evidence behind persistence: renewal behaviour, breadth of expansion, concentration, usage patterns and subsequent cohort results. A forecast should state those assumptions rather than copy the latest NRR indefinitely.

Avoid double counting in a revenue model

If the NRR input already captures price changes and upselling, adding the same pricing benefit again overstates revenue. Build the model either from an appropriately defined aggregate retention assumption or from its components, with a reconciliation between them.

Keep new-customer revenue outside the original cohort. Preserve currency and acquisition adjustments separately. An unavailable cohort breakdown should remain unavailable rather than become an AI-generated estimate presented as fact.

For the valuation context, read our SaaS modelling framework. For the general revenue reconciliation, use the volume, price and mix bridge.

Altys’s approach is to keep company research connected to sources and explicit definitions. Request access to evaluate a repeatable research workflow using a real company question. Retention belongs in that evidence record as a defined observation, not a shortcut to a stock recommendation.