CHURN IS DEAD
The 125% NRR That No CSM Ever Touched
11 min read · Revenue
Archive note: This issue predates the evidence ledger introduced in August 2026. Treat uncited benchmarks and examples as editorial analysis, not independently verified findings.
The 60-Second Version
In consumption pricing, most of your NRR is generated by the meter, not the motion, and the meter reports churn last.
Metered vs. Earned Retention splits your dollar NRR into the part that moved with usage no human influenced and the part a documented CS action changed, then watches usage velocity instead of dollars to catch the churn the meter is still hiding.
This week: pull your ten biggest "expansion" accounts and try to name, for each one, the specific CS action that caused the growth. The ones where you can't are your metered accounts, and you don't own them.
125 percent.
That is Snowflake's net revenue retention as reported in their most recent filings. Datadog sits around 120. These are the numbers the entire CS profession points at when it wants to prove that net retention is the most important metric in software, that a great NRR figure is the signature of a great customer success function.
Here is what almost nobody says out loud about those two numbers.
No QBR produced them. No success plan produced them. No CSM ran a play that moved Snowflake from 118 to 125. Those figures are what happens when a customer pays for compute and storage they consume, and they consume more of it every quarter because their own business is generating more data. The pricing model is doing the retention. The meter is the motion.
Snowflake could fire its entire CS organisation tomorrow and post a healthy NRR next quarter, because the number is largely a function of customer workload, not customer relationship. That is not a slur on their CS team. It is a description of consumption economics.
And if that is true at the top of the market, it is true in your portfolio too. The question is whether you can tell the difference between the retention you caused and the retention you happened to be standing next to when the meter ticked up.
Most CS orgs cannot. So let me show you the subtraction.
The account that expanded 118 percent and nobody touched
I want to walk you through a decomposition. The shape of this is drawn from consumption-priced enterprise accounts I work with. The numbers are changed and the account is unrecognisable on purpose. But the mechanics are exact, and you can run them against your own book on Monday.
An account renews at a consumption contract. Last year it burned 1.0 million in usage. This year the CS dashboard shows it landing at 1.18 million. A 118 percent expansion. The CSM books it as a win. It goes in the QBR deck. It goes in the board prep as evidence the account team is driving growth.
Now decompose it.
Of that 180K in growth, 155K arrived in a single ten-week window in the spring. Pull the usage logs and the reason is obvious: the customer ran a data migration off a legacy platform, and every record they moved passed through the meter as ingest. Nobody on the CS side planned it, influenced it, or even knew it was happening until the invoice moved. That is 155K of pure meter drift: dollars that moved with usage volume no human touched.
The remaining 25K came from a second business unit that the CSM personally onboarded after a workshop in Q2. There is a Slack thread, a success plan line item, and a named champion. That 25K is earned retention: growth with a documented CS action behind it.
So the honest read on this account is not 118 percent. It is roughly 92 percent metered plus 26 percent earned. The CS motion is responsible for the 26. The pricing model is responsible for the 92.
Watch what happens next.
The migration finishes in early summer. Ingest drops back to baseline. The 155K evaporates because it was always a one-time flow, not a new run rate. By the autumn the account is trending back toward 1.0 million and the "expansion" has quietly reversed. There was no red flag anywhere in the CS dashboard the entire time, because the dashboard watches dollars, and the dollars looked healthy right up until the meter dropped.
By the time the renewal conversation arrives, the trajectory is already decided. You are not intervening. You are witnessing.
Meter drift, and why the dollar reports churn last
Meter drift is the portion of your NRR that moves with usage volume no human influenced. Migrations. Seasonal batch jobs. A customer's own end-user growth. A one-off backfill. A model that got pointed at a bigger dataset.
Drift is not fake revenue. The money is real and it counts. The problem is that drift is exogenous. It arrives and departs on the customer's operational calendar, not on anything your team did. And it flatters you on the way up and ambushes you on the way down.
Here is the part the seat-based world never had to learn. In a seat contract, churn announces itself early. Seats go unassigned. Login frequency drops. Someone doesn't renew a block of licences and you see it coming a quarter out. The leading indicator and the dollar move roughly together.
In a consumption contract they come apart. Usage can fall for months while committed spend, prepaid credits, and annual floors hold the dollar number flat. The customer is quietly disengaging and the revenue line shows nothing. Then the commit resets, the credits burn down, and the number collapses all at once, long after the decision to leave was made.
That is the mechanical difference, and it has nothing to do with physics and everything to do with contract structure. The dollar is a lagging report of a decision the usage graph made months earlier.
Which means if you manage a consumption book by watching revenue, you are always managing it too late.
The consumption tell: watch the rate of change, not the dollar
The consumption tell is the signal you get from watching the rate of change in usage rather than the dollar output, because usage decelerates before revenue does.
This is where "watch velocity" stops being a slogan and becomes a column in your dashboard. Let me give you the actual instrument.
For every account, track three things weekly:
1. Trailing usage delta. This week's consumption versus the four-week trailing average, as a percentage. Not the dollar. The units: queries, ingest volume, compute-seconds, API calls, whatever your meter counts.
2. The direction of that delta over time. One down week is noise. What you want is the second-order signal: is the weekly delta getting more negative for three or four weeks running? A single account dropping from plus-two percent to minus-one to minus-four to minus-nine is decelerating structurally, even if the dollar hasn't flinched.
3. Distinct active teams or users. The number of separate teams touching the product. This is the one usage metric that is almost impossible to fake with a batch job, because it counts breadth of adoption, not volume of throughput. A migration spikes volume but not team count. A genuinely healthy account grows both.
Now set a flag. When an account shows a decelerating weekly usage delta for three consecutive weeks AND declining distinct-team count, it goes amber, regardless of what the dollar number says. If the dollar still looks like expansion while both of those are falling, it goes red, because that is the exact profile of my migration account: healthy revenue, collapsing engagement, meter about to drop.
That red flag would have fired on the 118 percent account in early summer. The dollar-based dashboard fired never.
The usage graph is the diagnosis. The dollar is the report you read at the funeral.
Earned retention, and why you have to name it account by account
Earned retention is the specific, named subset of accounts where a documented CS action changed the trajectory. Not the whole book. The subset.
The discipline is brutal and it is the whole point. For an account to count as earned, you have to be able to point to the artifact: the escalation you drove, the executive alignment you brokered, the second use case you stood up, the integration you unblocked, and tie it to the specific usage or dollar movement that followed. Slack thread, success plan line, dated action.
If you cannot name the action, it is not earned. It is metered. And most of your book, if you are honest, is metered.
This feels like an insult until you sit with it, and then it becomes the most useful thing your team has ever done. Because once you can separate the two, three things change immediately.
You stop staffing metered accounts like they need heroics. An account riding the meter upward on its own momentum does not need your best CSM running weekly touchpoints. It needs monitoring and a light touch. Your scarce human attention is being wasted admiring drift.
You start defending earned accounts like they are the actual product of your function, because they are. These are the accounts where your team demonstrably bent the curve. This is the ROI story that survives a CFO with a calculator.
And you get an honest denominator. When someone asks what CS contributed to NRR this year, you no longer wave at 118 percent and hope. You point at the earned column and say: this. We caused this specific slice. The rest was the pricing model, and you should thank product and finance for it, not us.
That sentence will make you unpopular for exactly one meeting and credible for every meeting after it.
The reversal window: the only place intervention still works
The reversal window is the lag between a usage collapse and the renewal conversation. It is the only stretch of time where a CS action can still change the outcome, and on most teams it is dead air.
Here is the timing that ruins consumption books. Usage starts decelerating in month one. The consumption tell would catch it there. The dollar number holds through months two, three, four because of commits and credits. The CSM, watching dollars, sees nothing. The renewal conversation opens in month six against a number that has finally started to sag, and now the CSM is scrambling to reverse a trajectory that has had five months to set.
The reversal window was months one through three. That is when the champion went quiet, the second use case stalled, the workload started migrating elsewhere. That is when a phone call, a re-scoping, a new use case could have re-engaged the account. By month six you are negotiating a smaller floor, not saving a customer.
The entire operational payoff of the consumption tell is that it hands you the reversal window. It moves the intervention from the renewal desk, where you are already losing, to the moment of deceleration, where you can still win.
A CS org that only wakes up when the dollar sags has structurally forfeited every reversal window it had.
The credit fight is the real story
Now the part nobody wants to run in a board deck.
If most of your NRR is metered, and metered NRR is a property of the pricing model, then the political economy of your CS team is built on a foundation you don't own. And two forces are pressing on that foundation at once.
From above, the CFO is starting to ask the decomposition question. Not "what's our NRR," but "what did CS cause." When the whole market posts elite NRR structurally, off consumption pricing, the headline number stops being a differentiator and starts being an expectation. The CFO who used to accept 118 percent as proof of a great CS team now wants to know how much of it would have happened with no CS team at all. If your answer is "all of it, honestly," you have just made the budget case against yourself.
From the side, expansion is migrating to sales. On large accounts, the moment growth is worth real money, the conversation gets pulled into an account-exec motion. Sales books the expansion and carries it to quota. CS keeps the gross retention risk and the churn defence. So the CSM is measured on an NRR number whose upside was booked by someone else and whose base is generated by the meter. Strong quarter, expansion booked by sales, meter did the rest, and the CS line reads as underperforming on a metric it was structurally cut out of.
This is the trap. You claimed NRR as your metric when it was flattering you. Now that it is being decomposed and redistributed, the claim is turning into the case for your own redundancy.
The way out is not to run from the number. It is to own the honest slice of it and defend that slice ferociously. Stop reporting blended NRR as a CS achievement. Report earned retention as your line, metered NRR as the pricing model's line, and reversal saves, accounts you caught in the window and pulled back from decline, as the leading proof that the earned column is real and repeatable.
That also fixes comp, which is the question everyone skips. You cannot pay a CSM a percentage of a number the meter generates. You can pay them on earned retention and reversal saves, which are the outcomes they actually produce. Comp the thing they cause, not the thing they witness.
The blended number was always going to get audited eventually. Better that you bring the decomposition to the room than that finance brings it to you.
What to do this week
1. Pull your ten largest expansion accounts. For each, try to name the specific dated CS action behind the growth. No artifact, no earned claim. Sort them into metered and earned.
2. Add three columns to your account view: trailing weekly usage delta, its direction over four weeks, and distinct-team count. Flag amber on three straight weeks of deceleration plus falling team count, red if the dollar still shows expansion.
3. Find one account that is currently green on dollars and amber on usage. That is your reversal window, open right now. Call the champion this week, not at renewal.
4. Build the two-line report for your next board or finance prep: earned retention (CS caused this) and metered NRR (the pricing model caused this). Bring it before someone asks.
5. Take one metered account off your best CSM's heavy-touch list. It is riding the meter without them. Redeploy that attention to an earned account or an open reversal window.
Go back to the 118 percent account. The one everyone celebrated in the spring and nobody defended in the autumn, because the dashboard never went red.
That account didn't churn on you at the renewal. It churned on you in March, in the usage graph, while the dollar number was still telling you a lovely story. The renewal was just the day you found out.
Work out which of your green accounts are already having that March. Then go earn the retention you keep taking credit for.
— Kuber
By Kuber Sethi · All issues · Subscribe