Recurring Revenue Metrics: MRR, ARR, NRR and GRR Explained

The four metrics

The first two are levels; the second two are rates. That distinction matters because levels can be grown by spending and rates cannot — which is why the second pair tells you far more about the business.

Calculating them without the common errors

1. 01
   MRR — contracted, recurring, normalised

   Only revenue under contract that recurs. Normalise annual contracts to a monthly figure. Exclude one-off implementation fees, professional services, and usage overages unless they genuinely repeat. Mixing these in is the single most common error and it makes the metric useless for forecasting.

2. 02
   ARR — pick one definition and hold it

   Either MRR × 12 or the sum of annualised contract values. Both are defensible; switching between them quarter to quarter is not. Write down which you use and where the boundary sits for multi-year deals.

3. 03
   GRR — cohort-based, expansion excluded

   Take a cohort's revenue at the start of a period, subtract churn and downgrades, and divide by the starting figure. Expansion is excluded entirely, so GRR can never exceed 100%. It is your product health measure.

4. 04
   NRR — same cohort, expansion included

   Same calculation with expansion added back. Above 100% means the cohort is worth more than it was a year ago without a single new logo. This is the metric investors weight most heavily.

GRR versus NRR: why you need both

NRR alone can conceal a real problem. A company with 115% NRR looks excellent. If its GRR is 78%, that headline number is being carried by expansion from a small number of accounts while the majority of customers leave.

The third row is the one that catches companies out, because the headline metric everyone reports looks strong right up until the concentration risk materialises.

Realistic benchmarks

Treat these as orientation rather than targets. The more useful comparison is your own trend by cohort — a business moving from 95% to 105% NRR is in a materially better position than one holding at 110% and drifting down.

Committed versus booked

For planning purposes, booked ARR overstates what you actually have. Two adjustments make the number usable:

• Exclude accounts in notice. Revenue from a customer who has given notice is not committed revenue, however long it remains on the books.
• Flag at-risk renewals separately. Accounts with a health score below threshold or a renewal inside 90 days with no engagement should be visible as a distinct line, not blended into the total.

Committed ARR — booked, minus notice, with at-risk flagged — is the number worth planning against. It is usually 3–8% below booked ARR, and the gap itself is a useful early warning indicator when it widens.

Instrumenting them properly

These metrics are only as good as the contract data behind them, which is where most implementations break. Three requirements:

• Contract terms live in one system, with start date, end date, value, and billing frequency as structured fields rather than in a PDF or a note.
• Changes are events, not overwrites. An upgrade should create a record of the change, not replace the old value — otherwise you cannot reconstruct a cohort's history.
• One definition of churn date. Notice date, end-of-term date, or last-payment date. Pick one, write it down, and use it everywhere.

The second requirement is the one most commonly missed and the hardest to retrofit, because once you have overwritten the history it is gone. Getting it right early is a systems design decision, and it is why retention reporting is usually a RevOps deliverable rather than a finance one. The subscription modelling this depends on is covered in RevOps for SaaS companies.