Net revenue retention (NRR): how to calculate and read it

Net revenue retention (NRR) measures how much recurring revenue a company keeps and grows from the customers it already had at the start of a period: NRR = (beginning ARR + expansion - contraction - churn) / beginning ARR. Gross revenue retention (GRR) uses the same cohort but counts only the losses. NRR is one of the first SaaS metrics an investor asks about, and one of the easiest to get wrong: two companies can report the same NRR and mean very different things, because one counted new customers, one measured over a month and annualized it, and one quietly left out downgrades.

ARR, NRR and GRR: what each means

• ARR (annual recurring revenue) is contracted recurring subscription revenue, annualized, and it is the base both retention metrics are measured against.
• NRR (net revenue retention) is the ARR the starting customers produce at the end of the period, after expansion, contraction and churn, divided by their beginning ARR, so it can exceed 100%.
• GRR (gross revenue retention) is the same ratio with only contraction and churn counted and no credit for expansion, so it cannot exceed 100%.

ARR is a dollar figure; NRR and GRR are percentages calculated from it for the same cohort of customers, and the gap between them comes from expansion.

ARR vs NRR vs GRR compared

Illustrative figures, not client data.

ARR vs GRR. ARR is the size of the base, and it rises with new customers; GRR is a rate that ignores new customers and expansion and shows only how much starting revenue was kept. ARR can grow quickly while GRR falls, if new sales cover rising churn, which is why investors ask for both.

GRR vs NRR. Both use the same starting cohort and the same contraction and churn. The only difference is expansion: NRR adds it and GRR does not, so NRR minus GRR equals expansion as a share of beginning ARR. In the example below that gap is 17.5 points (70 / 400). A wide gap means growth depends on upsell; a narrow gap with a high GRR means a sticky base with little expansion.

What NRR and GRR measure

Both metrics look only at customers the company already had at the start of the period. Revenue from customers who signed during the period is excluded from both.

• Net revenue retention (NRR) includes everything that happened to that starting revenue: expansion, contraction and churn. It shows whether the existing base grows or shrinks on its own.
• Gross revenue retention (GRR) includes only the losses: contraction and churn. It shows how much of the starting revenue the company kept, before any upsell.

Read together, they separate how well the product holds customers from how well the company grows them.

The formulas

Start with the cohort: every customer with ARR at the beginning of the period. Then, for that cohort only:

NRR = (beginning ARR + expansion - contraction - churn) / beginning ARR

GRR = (beginning ARR - contraction - churn) / beginning ARR

Expansion covers upsells, added seats, price increases and cross-sells to existing customers. Contraction covers downgrades, seat reductions and discounts on renewal. Churn is the full ARR of customers who left.

GRR cannot exceed 100%. It has no term that adds revenue, so the best possible result is keeping everything. A GRR above 100% means expansion has leaked into the calculation. NRR can exceed 100% when expansion outweighs contraction and churn.

One detail matters for GRR. If a customer upgrades in one product and downgrades in another, measure the customer's net change. A customer who ends the period above their starting ARR counts at their starting ARR for GRR, never above it.

NRR, NDR and other names

Net revenue retention and net dollar retention (NDR) are the same metric under two names. Some companies also say net retention rate or net expansion rate. Gross dollar retention is the same as GRR.

Two companies can both say "NDR" and calculate it differently, so the definition should always travel with the number. Public filings show how far definitions differ: Datadog's annual report describes a trailing 12-month dollar-based net retention rate built from ARR, while Snowflake's describes a net revenue retention rate built from product revenue over a two-year measurement period.

Trailing twelve months or monthly annualized

Trailing twelve months

Take the customers who had ARR exactly twelve months ago. Compare their ARR then with their ARR today. Customers who signed in the last twelve months are excluded entirely. This is the most common method for board and investor reporting, because it covers a full renewal cycle for annual contracts and smooths out lumpy months.

Monthly, annualized

Measure retention for one month, using the cohort at the start of that month, then annualize it. This responds faster to change, which helps for internal monitoring. But the annualizing step is a choice. Compounding a monthly NRR of 99.5% gives about 94.2% for the year (0.995 to the 12th power). Multiplying the 0.5% monthly loss by twelve gives 94.0%. Both are defensible. They are not interchangeable.

Monthly figures also carry more noise. A single large renewal or downgrade can swing one month, and annualizing that month magnifies it.

State the method every time

Whichever method the company uses, write it down and use it every period: the measurement window, how the cohort is selected, and how any monthly figure is annualized. Without its method, a retention number cannot be compared with anything, even the company's prior quarters.

Build the ARR waterfall first

Retention should come out of the ARR waterfall, not be calculated separately. The waterfall explains how ARR moved from the start of a period to the end:

Beginning ARR + new + expansion - contraction - churn = ending ARR

To build one that supports NRR and GRR:

1. Write the ARR definition. Contracted recurring subscription revenue, annualized. Decide and document how to treat one-time fees, services, usage, multi-year price steps and free months. Most companies exclude anything not recurring and contracted.
2. Build ARR by customer by month. One row per customer, one column per month, from the contract or billing system. This is the source for everything else.
3. Classify every month-over-month change. Each change for each customer falls into exactly one bucket: new (first month with ARR), expansion (increase), contraction (decrease but still a customer), or churn (falls to zero).
4. Set a rule for returning customers. A customer who churns and comes back can be treated as new or as a reactivation. Pick one and apply it consistently.
5. Tie it out. Beginning ARR plus the four movements must equal ending ARR, every month. Ending ARR should reconcile to the contract list and, roughly, to recognized subscription revenue in the books.

With customer-level data, the NRR cohort is simply the customers with ARR at the start of the window, and their movements can be summed directly. The same waterfall supplies the net new ARR behind the burn multiple.

A worked example

A small company starts the year with five customers. One new customer signs during the year. All figures are ARR in $K.

Illustrative figures, not client data.

The ARR waterfall for the year:

Illustrative figures, not client data.

Retention for the starting cohort of customers A to E:

Illustrative figures, not client data.

The existing base grew slightly on its own, but GRR shows nearly 14% of starting ARR was lost to churn and downgrades, covered by two customers' expansion. The next question a reader should ask: how dependent is NRR on a few accounts?

The example also shows two of the mistakes below. Including customer F would give 465 / 400, or about 116%, which is ARR growth, not retention. And logo churn (20%) and dollar churn (10%) tell different stories, because the customer lost was one of the smallest.

Common calculation mistakes

• Including new-customer revenue. That measures ARR growth; NRR uses the starting cohort only.
• Mixing logo churn and dollar churn. NRR and GRR use dollars, not customer counts; never blend the two.
• Inconsistent ARR definitions. If services or usage were in ARR last quarter and out this quarter, retention moves for reasons that have nothing to do with customers. Restate prior periods when a definition changes.
• Excluding downgrades. Counting only full churn overstates both NRR and GRR. Contraction belongs in both.
• Cohort timing errors. Customers who signed partway through the window should not be in the cohort. A customer who signed eleven months ago and expanded last month is new revenue, not retention, on a trailing twelve month basis.
• Counting bookings before they start. A signed upsell that starts next quarter is not ARR yet.

How to improve each

The levers are different, which is one reason to report both.

Improving GRR

GRR improves when fewer customers leave and fewer cut back.

• Find where churn concentrates: by segment, plan, contract size, onboarding cohort or sales channel. Losses are rarely spread evenly.
• Watch leading signals such as usage, support volume and renewal conversations well before the renewal date.
• Review discounting at renewal. A renewal at a lower price is contraction, even if it feels like a save.
• Consider contract structure: annual terms and clear renewal processes reduce surprise churn.

Improving NRR

NRR improves through GRR plus expansion.

• Make sure pricing has room to grow with the customer: seats, usage tiers, modules or editions.
• Give someone ownership of expansion, whether account management or customer success, with a target measured on the same definitions.
• Track expansion by customer, not just in total. Broad expansion across many accounts is more durable than a few large upsells.

How an investor reads your retention

Decipher sets up a company's finance the way an investor or acquirer will read it. On retention, the same checks come up in almost every diligence process, from a funding round to a quality of earnings review before a sale.

• Is the definition written down and stable? An investor will ask how ARR, expansion, contraction and churn are defined, which method sets the period, and whether any of it changed. A metric definitions page in the data room answers this before it is asked.
• Is there a cohort table? A single company-wide NRR hides a lot. Investors want retention by cohort, usually grouped by the quarter or year customers signed, to see whether newer customers retain as well as older ones.
• Does it reconcile to the waterfall and the books? The expansion, contraction and churn behind NRR should be the same figures as the ARR waterfall for the same period, and ending ARR should reconcile to customer contracts and, at a high level, to recognized subscription revenue. An unexplained gap undermines the retention figure built on top.
• Can they recompute it? Investors often ask for customer-level ARR by month so they can calculate retention themselves. If their number differs from the company's, the company's credibility takes the hit.
• Is expansion concentrated? Expect a look at how much of the expansion came from the largest few customers, and what NRR would be without them.
• Is GRR reported alongside NRR? A strong NRR shown alone invites the question of what GRR is. Showing both up front reads as confidence.

None of this requires sophisticated tooling, only customer-level ARR data and written definitions applied the same way every month. Decipher builds this as part of its FP&A services. To see where a company's finance function stands today, the finance scorecard takes about two minutes.