ChartMogul ran 35,512 cohort-quarters across 3,331 companies to test whether the standard LTV formula predicts anything. Thomas Anastaselos wrote it, with research guidance from Jason Cohen. It's careful work, the method is sound, and the data is the most useful thing published on this in a while.
I have no argument with the calculation. That isn't where this goes.
The formula is accurate against the data you have. The data, and the revenue it represents, is just incomplete. That's the real blindspot.
The formula is fine
LTV as ARPA divided by churn does what it says. It's been computed that way for twenty years and it can go on being computed that way.
The report finds it unreliable in the usual ways, and those findings are real: 28.3% of cohorts off by more than 50%, blended ARPA carrying years of other customers' expansion, logo churn missing contraction entirely. All true. All worth fixing. None of that's my point, and I'd rather not spend the argument there, because arguing about the model implies the model is what's between you and the number.
It isn't.
What the number is actually measuring
Here's the finding that should have been on the cover.
45.6% of current MRR from customers who survived from 2020 to 2021 came from expansion after signup, not from their original subscriptions.
Nearly half the revenue in a mature book was never sold at signature. It arrived later, from customers who were already there.
Now ask what that 45.6% is a measurement of.
Start one layer under it. Most companies can't say what a customer should be worth over time. Not the invoice, not the renewal, but the whole of what that account was ever going to be worth given their actual progress, their ascension path, and the gates they pass on the way. If nobody can state that number, nothing gets built to go collect it, and what arrives is whatever arrives.
It's a measurement of what got collected. Not what was available. In those 3,331 companies, over those five years, expansion was not a motion. Nobody owned it, nothing triggered it, no forecast contained it. Somebody noticed an account looked healthy, or a renewal came up and a rep asked a question, or a customer went and asked for more on their own.
So 45.6% is the rate at which revenue gets collected in spite of the company rather than because of it. There was no expansion motion in place. Whatever they got, they got.
LTV has never measured what a customer was worth. It measures what you captured. Those get treated as the same number, and in most companies they are, but not because they're the same thing. Because the difference was never collected, so it never showed up to be counted.
There are two of these numbers
One has been computed 35,512 times, carefully, and we now know a lot about how reliable it is.
Most companies have never computed the other one.
It's the LTV those same cohorts would have produced if the companies holding them had run a real expansion motion. Not a better forecast of what happened. A different result, because different things would have been sold, at different moments, to customers who were ready for them.
That number runs substantially higher, and how much higher is specific to the book. It moves with how much was bundled away at signature, how many gates the lifecycle actually has, and how far the accounts have already drifted past them.
The comparison worth running is captured against available. It's runnable. It just isn't run in most places, which is why a study of 3,331 companies reads the way this one does.
What has to exist before that number can
Latent revenue has to be identified first, and it has a definition: the revenue associated with expansion opportunities tied to readiness gates across the customer lifecycle.
Not a vague sense that accounts have room. A specific inventory. This customer, at this gate, is ready for this thing, and here is the evidence they arrived. That's an object you can count, forecast against, and be wrong about in a way you can check.
Then it has to be orchestrated. A repeatable motion that moves when readiness appears rather than when the renewal calendar does.
Do those two things and you generate data most companies don't have. Skip them and every study of LTV, however careful, is a study of what showed up anyway.
Data can't contain what nobody operationalized
This is the part no amount of analysis fixes, and it isn't a criticism of the analysis.
You can compute anything you like against five years of historical cohorts. What you can't do is find a pattern the world never produced. If a practice was never identified and never built into a repeatable motion, there's no signal from it in the record, and no method recovers a signal that was never generated.
Three things would've had to exist for that dataset to contain the number I care about.
Latent revenue discovery. Going account by account and establishing what each customer is ready for that nobody has named. Present, real, capable of emerging, not yet visible. The customers exist now, their success is accumulating now, their readiness is building now, whether anyone is watching or not. Nothing about it is hypothetical except the collecting.
Strategic unbundling. Not putting everything into the initial deal at a discount to close the quarter. Holding back the thing the customer can't use yet, so it can be presented at the moment it solves a problem they can feel, and priced then. The same item is worth a different amount depending on when it shows up.
Orchestration on a cadence. A repeatable motion that watches for readiness and moves when it appears, rather than when the renewal calendar arrives. Milestones, events, changes on their side, what they actually did, where they sit in their lifecycle.
Almost none of that was running in that dataset. So the analysis is complete and correct about a world where hardly anybody did any of it.
Which is why the report's fixes stop where they do
Their recommendations are calibration. Discount e-commerce 10 to 15%. Haircut about 20% above $30M ARR. Replace blended ARPA with signup-time cohort revenue.
The last one is structurally right and costs nothing, so take it.
The first two make a forecast less wrong about an unmanaged process. That's a smaller ambition than managing the process, and it's the whole difference between treating LTV as weather and treating it as an outcome you set.
One mechanism they found and didn't name
Cheapest quartile by signup MRR survived at 74%. Most expensive survived at 45%. They report it as an inversion of conventional wisdom and leave it there.
Here's a cause. The expensive customers were oversold.
The deal that produced the high signup ARPA is the same deal that got everything bundled in to close the quarter. The customer bought capacity they couldn't use yet, at a price set before they had any way to judge it, and the ceiling and the exit were both fixed at signature. High ARPA and short life aren't two findings. They're one decision showing up twice.
That isn't a churn insight. It's a sequencing one, and it's exactly the revenue that strategic unbundling would have kept.
The blindspot
It isn't in the model. The model is fine.
It's the revenue that exists, right now, in accounts you already have, that nobody has identified and nobody has built a way to collect. Sitting there the whole time, never collected. It doesn't appear in your LTV because it never appeared in your bank account, and it never got there because nobody asked for it at the moment it was worth asking.
Take their cohort-level fix, which is genuinely good. Then stop treating 45.6% as a benchmark. It's a floor, and it's the floor for companies that weren't even trying.
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.