Net revenue retention gets filed as a customer success metric. It sits on the retention slide, it gets discussed after churn, and when it disappoints the conversation is about the post-sale team.
That's the wrong department, and usually the wrong year.
What actually moves the number
Three things, and only three: retention, contraction, and expansion. Customers who stay, customers who shrink, customers who buy more. Everything else is arithmetic on top of those.
Now ask when each one was decided.
Retention. A meaningful share of churn was set at signature, by a promise nobody could keep or a fit nobody checked. I've made that argument at length: the sales call sets your churn date, often months before anybody post-sale meets the customer.
Contraction. Somebody bought thirty seats because a rep sized the deal for the quarter rather than for the customer, or took a bundle they'd use a third of. The downgrade at renewal isn't a change of heart. It's the account settling to the size it should have been sold at.
Expansion. This one is the clearest. If everything got sold at once, there is nothing left to present when the customer succeeds, and a customer doing well produces exactly the same revenue as one barely hanging on. That ceiling was set at signature and nothing post-sale removes it.
So the three inputs to NRR were all substantially determined before the customer became a customer.
Which makes NRR a lagging indicator of ARR quality
I've written about the difference between increasing ARR and improving it. Two companies can post identical ARR with completely different revenue underneath: one discounted into logos it will spend next year defending, the other expanded existing accounts at full price with a clear next step.
The day you book it, those look the same. ARR cannot show you the difference and it never will, because ARR is a total and quality is a property of the parts.
NRR is where the difference finally surfaces. Low-quality ARR shows up eighteen months later as more churn, more contraction, and less expansion, which is to say it shows up as NRR.
That's the whole relationship:
ARR quality is the leading indicator. NRR is the lagging one. What you did at signature is what you read about later.
Blended NRR hides it by construction
One number, rolled up across every customer you have, is a ratio of aggregates. Ratios of aggregates conceal divergence. That's not a flaw in how anyone reports it, it's what averaging does.
So a company whose 2023 vintage stopped expanding three quarters ago can post a blended NRR that's flat or improving, because newer cohorts are still in their expansion window and carrying the average. The board sees one number holding and nobody asks a second question.
Cohort NRR by account age fixes that. Not as a dashboard preference. Because the thing you need to know is whether expansion stops, and at what age, and a blended number is structurally incapable of telling you.
The follow-up question is the one that matters: is the stall point getting earlier? A vintage that stopped expanding at month thirty is a fact. Three vintages stalling at thirty, then twenty-six, then twenty-two is a trend, and it's telling you something got worse at the front of the business.
The part that makes it usable
Everything above is diagnosis. Here's the move.
Draw a line. Pick a date. Change what happens at signature: what gets held back, what condition gets named, what gets written down and passed on.
Then watch the cohorts on either side of that line diverge.
That's not a report anymore, it's an experiment, and it's the only way I know to prove a sales-side change actually worked. If cohorts after the line retain better, contract less, and expand more than cohorts before it, the change took. If they look the same, you changed something that didn't matter, and you'd rather find that out from the data than from a story about how it feels different now.
And why almost nobody runs it
The feedback loop is eighteen to twenty-four months.
That's longer than a planning cycle, longer than most tenures in the seat, and far longer than anybody's patience. So the change gets made, nothing visibly happens for six quarters, and attention moves on well before the evidence arrives.
Which is exactly why you need both ends of it.
The leading indicator is the sales call itself. What got promised, what got given away, what condition got named, what the customer said once that nobody wrote down. You can read that this week, and it tells you what your NRR is going to look like long before NRR does.
The lagging confirmation is cohort NRR. Slower, but it's the thing that proves you were right.
Run only the lagging one and you're steering a ship by its wake. Run only the leading one and you're guessing that your reading of a call predicts anything. Run both and you have a loop: change the input, read it immediately, confirm it eventually.
Lincoln Murphy formally named and popularized Customer Success starting in 2010 and has spent 15 years connecting it to expansion revenue and commercial outcomes. Read The Premise.