What Retention Is Made Of
Most of the number was decided before your team met the customer. The part you can move is smaller than the part you're measured on, and naming the difference is the job.
Most of the number was decided before your team met the customer. The part you can move is smaller than the part you're measured on, and naming the difference is the job.
You'll own the retention number.
It gets said in the third conversation, usually by the founder, and both people nod. The founder means that a number which has been nobody's in particular is now somebody's, and the somebody is you. The candidate means they'll do the work that makes customers stay. Those aren't the same sentence. The distance between them is where a lot of customer success leadership quietly ends, about seven quarters later, with a number that moved the wrong way and an explanation nobody has patience for.
The way out isn't to negotiate the target down. It's to take the number apart before agreeing to it, because the argument about who owns retention is unwinnable in the abstract and nearly trivial once the number is in pieces.
A retention number isn't a measurement of what one team did last year. It's an aggregate of decisions made at different times, by different people, and most of them were made before anyone in customer success had met the account.
There are four places a renewal outcome actually gets decided.
At the point of sale. Who was sold to, against which use case, at what discount, with what promised. An account sold to a champion with no budget authority, into a workflow the product serves badly, is a loss that got booked as a win eighteen months before it shows up in your number.
In the first ninety days. Whether the thing got implemented properly and whether anybody got value from it. This one is shared, and it's the only one of the four where the sharing is usually honest.
In the customer's own business. They got acquired, the sponsor's function was eliminated, the budget was cut by somebody who has never heard of you. Genuinely outside the building.
In the renewal year, by whether the value was legible. The product worked, the outcomes happened, and at budget time nobody inside the customer could say what it had been for. This is the delivery layer, and it's the piece I've argued elsewhere doesn't compress and doesn't automate.
Four authors, one number, and the number carries one name.
Invented numbers, real proportions. A company at $20M ARR. Over the year it loses $3.0M to churn and contraction and adds $2.4M in expansion, so gross retention is 85% and net retention is 97%.1Gross retention is what you kept. Net adds expansion back on top, which is why a healthy net number can sit comfortably over a book that's bleeding, and why a leader handed only the net number should ask for both before agreeing to anything.
Ninety-seven is the number on the board slide. Three million is the number that decides whether anyone still works here in two years.
Take the $3.0M apart by when each loss was actually decided:
The bottom row is the one a customer success team is usually measured on. It's a tenth of the loss.
Run the same exercise on a real book and the proportions move. The shape doesn't: the two largest blocks are always the two that were finished before the renewal year began.
A retention number is a report on decisions, and most of the decisions in it are old.
Here's where the decomposition gets misread, including by people who like it.
Where a loss was decided isn't the same as where it could have been prevented, and it's nowhere near the same as what you can change now. The four authors tell you where to send the evidence. They don't tell you how much of the number is movable, and treating attribution as a defense is how a customer success leader talks themselves into a small job.
The honest position is narrower and more useful. Every slice is movable. They just run on different clocks, and only one of them runs fast enough to land inside your first year.
That's the sentence to say in the interview, and it's the one that separates a leader who has owned a number from one who has only been assigned one.
Every one of these can be started in month one. Only the first one pays inside the year you're being judged on.
This is the part that gets missed, and it isn't a subtlety. When a retention number moves, the cause is usually two years old. So the number you inherit in January is a report on the previous leader's sales qualification and the previous year's onboarding, and the number you hand over is mostly a report on yours.
Which means the first-year plan has an obvious shape once you accept it. Move legibility, because it's the only slice that pays this year. Fix onboarding, because it pays next year and next year arrives. Start the qualification argument immediately, knowing you're building it for a successor or for your own third year. And route the product evidence without pretending it's a lever you hold.
The slices that aren't yours are still the ones you see before anybody else does. Customer success watches fit failure, implementation failure, and product failure at the account level, in real time, a year before any of it reaches a board deck. That's the real asset, and most teams waste it by filing it as commentary.
This is the delivery-layer problem again, pointed inward. A finding about qualification is worth exactly as much as its odds of being repeated by the person who owns qualification, and the content of the finding isn't what decides those odds.
The second list is smaller, and it's close to the whole of what a first year buys. A leader who won't say that plainly gets judged against the first list regardless, having never named the difference.
The hard case is qualification, because the person who owns it is the CRO, and evidence about it looks like blame no matter how it's phrased. So it can't be routed as a story about a deal or a rep. It goes as arithmetic, quarterly, unattributed and boring: here's the renewal rate of accounts sold into this use case against that one, here's the same cut by whether the economic buyer was in the room. Nobody has to lose an argument for the number to change the behavior, and a fact that needs somebody to lose an argument doesn't travel.
Six hundred thousand of that three million was genuinely outside the building. Acquisitions, budget cuts, a sponsor whose whole function was eliminated. Real, and unwinnable.
It's also the number every team I've seen inflates, and it inflates for an honest reason. The account team was there. They watched it happen. From inside a losing account the terminal cause is vivid and the earlier one is invisible, so the loss gets filed under the last thing that happened to it.
The fastest real win available in a first quarter is to re-read last year's losses with the excuse removed. Take every account filed under budget cut or acquisition and ask two questions of each. Was there a second relationship when the champion left? Was anybody able to say what the thing was for, in a sentence, without help, at month nine?
A meaningful share of the weather turns out to have been a delivery failure that a real event later put out of its misery, and that share isn't weather. It's your slice, and it was already your slice before anyone re-read anything. It's the same concealment I described in coverage math: a loss that had no metric attached while it was happening gets explained, afterwards, by whatever did.
I wouldn't present that reading to a board in the first quarter. I'd use it to pick the twenty accounts to work on, and let the number make the argument later.
You don't get to decline accountability for the whole number. Nor should you. A leader who spends the first meeting explaining which parts aren't theirs has told the room they'll be doing that again in a year, at a worse moment.
Take the whole number. Then say the three sentences.
Here's what it's made of, on your book, not in general. Losses by when the outcome was decided, from your own list, done by hand in the first month. It's a week of work, and almost nobody arrives having done it.
Here's the part that moves this year, and it's the smaller part. Legibility and the risk list. Name the figure you expect and name it low, because the credibility of everything else you say for two years is priced off this one estimate.
Here's what I'm starting now that pays after the review that judges me. Onboarding, qualification, the product evidence. Say plainly that these land late. A founder who understands that is a founder who'll still be sponsoring the work in month fourteen. One who doesn't is telling you something useful about the job while you can still decline it.
The claim everything rests on is the clock speeds, and it's the one I'm least able to prove. I've watched qualification changes take two years to appear in a number, and I'm generalizing from a small number of books, all of them annual contracts in the enterprise. Shorten the contract and the whole argument compresses. On monthly terms every clock runs faster, the lag between a fix and its evidence collapses, and a leader really can move more of the number inside a year. Most of what I'm describing is a statement about contract length wearing a statement about organizations.
The stronger objection is that the decomposition is unfalsifiable in practice. Attributing a loss to a qualification call eighteen months earlier is a judgment call made by people with an interest in the answer, and the same account can be filed under three of the four authors by three honest people. I think it survives that, because the exercise is comparative rather than forensic. You're not proving where one account died, you're looking for the block that's consistently large. But it's a real weakness, and anybody using this to settle a specific argument about a specific account is using it for something it can't do.
There's also the version where the whole frame is a way of being interesting about a number that mostly follows the product. If the product is good enough, retention is high and none of this decides anything. That's true more often than the profession likes to admit. It stops being true at exactly the size where somebody first gets hired to own the number, which is the situation this is written for.
If you try one thing from this, take last year's losses, one line per account, and write down when each outcome was actually decided rather than when it was recorded. It takes a week, and it doesn't need anyone's permission. Most books have never been read this way, the largest block is almost never the one the team talks about, and the answer changes what you should spend your first year doing more than any target anybody hands you.
If you're about to accept one of those targets and want to talk through the decomposition first, reach out. I'd be glad to.