Net Revenue Retention (NRR)
Net revenue retention (NRR) measures how much recurring revenue a fixed group of existing customers produces in this period compared with what the same group produced a period earlier, counting upgrades, downgrades and cancellations, and excluding anyone new. Above 100% means the customers you already have grow revenue on their own; below 100% means every new sale has to refill a leaking bucket before it adds anything.
The formula, and the version of it that lies
Take a cohort's monthly recurring revenue at the start of a period. Add the expansion that cohort produced through upgrades, added seats or usage above their plan; subtract contraction from downgrades; subtract the revenue lost to cancellations. Divide by where the cohort started. New customers acquired during the period are excluded — that exclusion is the whole point, because including them measures growth and calls it retention. Gross revenue retention (GRR) is the identical calculation with expansion set to zero, which is why GRR can never exceed 100%.
The gap between those two numbers is where the information lives, and quoting NRR alone is how the metric misleads. A business shedding a fifth of its revenue to churn can still post NRR comfortably above 100% if a handful of large accounts happen to be expanding, and that reading looks like health right up to the quarter those accounts stop expanding. GRR tells you how leaky the bucket is; NRR tells you whether expansion is currently patching it faster than it leaks. One number without the other is a sentence with the verb removed.
Why founders compute it years too early
NRR needs a denominator large enough that no single customer can move it. With a dozen paying accounts, one upgrade swings the figure by tens of points and one cancellation swings it back, so what you are holding is an anecdote with a percent sign attached. It also needs enough elapsed time to mean anything: a monthly NRR read on a product people evaluate over a quarter is largely a description of your billing cycle. Early on, the honest substitutes are simpler and more actionable — how many accounts renewed at all, and a written reason for every single cancellation, gathered by asking rather than inferred from a dashboard.
There is also a structural trap in comparing your NRR to a published benchmark. Expansion is a property of your pricing model before it is a property of your product: a seat-based or usage-based plan expands on its own as a customer grows, while a flat self-serve plan has almost no expansion mechanism at all, so its NRR is capped near its GRR no matter how much customers love it. Comparing the two is comparing pricing models. If your NRR looks weak against an enterprise figure, the first thing to check is whether your plan structure even permits the number you are chasing, and the second is whether reducing churn is the cheaper lever.
What we track instead, and why we publish no NRR figure
We do not report an NRR number for AgentCeres — the AI Growth Officer at agentceres.com — and the reason is the denominator problem above rather than modesty. Our paid base is small enough that one customer moving up a plan would shift the figure by tens of points, which would make it a statement about that one customer wearing the costume of a company metric. The ingredients genuinely exist in our billing: plan upgrades, credit top-ups and usage above a plan's included allowance are all expansion in the strict sense. We could compute something. It would just be a number whose error bars are wider than the thing it claims to measure.
So we watch leading behaviour instead: whether a customer comes back on a second and a third day, and whether they connect a real data source rather than only talking to their agent once. Those are countable at small numbers in a way NRR is not. The most useful thing they have told us is uncomfortable and has nothing to do with the product — where a signup came from predicts whether they return better than anything we do after they arrive, which meant the fix for a weak retention number was upstream in acquisition, not in the onboarding flow we were busy rewriting. If you are choosing a first retention metric, activation survives small samples; NRR does not.
FAQ
- What counts as a good NRR?
- 100% is the only threshold with an unambiguous meaning: above it, the customers you already have fund growth by themselves; below it, new sales are partly replacing revenue you already had. Beyond that line, published benchmarks are hard to borrow, because they mostly reflect pricing model and market segment rather than product quality — a seat-based enterprise vendor expands as its customers hire, while a flat self-serve plan has almost no expansion path. Compare your NRR to your own previous quarters first, and to companies with your pricing shape second.
- Does NRR include new customers?
- No, and this is the single most common error in the calculation. NRR is measured on a fixed cohort — the customers who were already there at the start of the period — and follows only what happened to that group. Add new sales into the numerator and the metric will comfortably exceed 100% while your existing customers quietly leave, which is precisely the failure it was invented to expose.
- Should I report NRR or GRR?
- Both, together. GRR is the floor: it shows how much of your revenue survives without help from upgrades. NRR is the ceiling: it shows what expansion adds on top. Reported as a pair they are hard to misread, because a wide gap immediately raises the useful question of whether expansion is broad across the base or concentrated in a few accounts.
- Can NRR be above 100% while the business is in trouble?
- Yes, and concentration is how. If a small number of large accounts are expanding fast enough, they can mask heavy churn across everyone else, and the aggregate reads as health while your actual customer count falls. The check is to look at NRR alongside GRR and logo retention, and to compute NRR with your largest account excluded. If the number changes character when one customer leaves the calculation, that customer is your retention story.
An AI growth team that runs this for you
AgentCeres is a managed AI marketing team — you approve what ships. 14-day free trial, from $39/month.