पाठशाला Pathshala · हिसाब Hisāb, Unit economics · Lesson 14 · Build
LTV, computed honestly
The lifetime value in most decks is a price multiplied by a lifetime nobody has observed. Build it instead from the retention you have actually seen and the margin you actually keep, and stop where the data stops.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
Most lifetime values in Indian pitch decks are written by multiplication: a monthly price, a lifetime of five years, a ratio to CAC that comes out somewhere above ten. The company that wrote it has usually been selling for fourteen months. Lifetime value is a real number and a useful one, but only when it is built from what customers have done rather than from what the founders hope they will do.
This lesson names the three errors in the deck figure, builds the honest version from a cohort’s survival curve, works a subscription app through both and sets a rule for how long a lifetime a company may claim. The figure runs your own curve.
The deck number and its three errors
Andreessen Horowitz’s 16 Startup Metrics defines lifetime value as the present value of the future net profit from a customer over the duration of the relationship, and names the common mistake: estimating it as the present value of revenue. That is the first error. A ₹499 subscription that costs ₹150 a month to serve, collect and support leaves ₹349, and only the ₹349 can repay acquisition. Use [contribution](/library/contribution-margin-first-number-to-know), not price.
The second error is the lifetime. Five years is a guess unless the company has customers who have stayed five years. The textbook shortcut, average lifespan equals one divided by monthly churn, is no better if the churn fed into it is the wrong one. Feed it the churn of the first month and lifetime looks short. Feed it the churn of customers who have already stayed a year and it looks long, because it applies the loyalty of survivors to everyone who signed up.
The third error is time. A rupee of contribution in month thirty is worth less than a rupee today, both because money costs something and because the forecast is less certain the further it runs. The deck figure discounts nothing.
Start from the curve you have observed
The honest method begins with a cohort grid: customers grouped by the month they started, and for each cohort the share still paying one, two, three months later. The [cohort lesson](/library/cohort-analysis-for-founders-not-analysts) builds it from a payments export, and the [retention curves lesson](/library/retention-curves-and-flattening-test) shows how to tell whether it flattens. Average the cohorts old enough to be reliable and you have a survival curve: 100 per cent in month one, falling fast at first and then slowly.
That shape is not customers growing more loyal. Peter Fader and Bruce Hardie’s paper How to Project Customer Retention explains it as a sorting effect in a mixed population: the customers likely to leave leave early, so the ones who remain have lower churn from the start. Any lifetime calculation that uses one churn rate for everybody is wrong in a predictable way. The fix is to treat the curve as two groups, the many who leave fast and the floor who settle, and to project only the floor beyond the months you have observed, at the tail churn the floor actually shows.
Then multiply each month’s survival by contribution per customer, discount it back at a monthly rate derived from your annual cost of capital, and add the months up. The sum is lifetime value per customer acquired. Stop it at a horizon you can defend, which the section after the figure sets.
A subscription app, worked
A fitness app in Pune charges ₹499 a month and keeps 70 per cent as contribution after payment fees, servers and support: ₹349. It pays ₹1,800 to acquire a subscriber. Its deck says the average subscriber stays five years and puts lifetime value at ₹29,940, more than sixteen times CAC.
Its cohorts say something else. Of every hundred who pay in month one, fifty-five pay in month two. The curve falls to under twenty-five by month six and then levels off, with the settled subscribers leaving at about 2 per cent a month. The company has twelve months of data. Sum contribution along that curve to month twelve and lifetime value is about ₹1,300, under one times CAC. Extend the settled quarter at 2 per cent a month to thirty-six months, discount at 15 per cent a year, and it reaches about ₹2,340: 1.3 times CAC. The textbook formula with the 2 per cent tail churn gives about ₹17,500, which is the value of a settled subscriber wrongly applied to every one of the hundred.
Nothing about the product changed between ₹29,940 and ₹2,340. The second number says the company does not yet earn back its CAC inside a year and earns it back only modestly over three, and that the lever is not more acquisition but a higher floor: more of the hundred settling.
The figure opens on the Pune app. Raise the floor from 25 to 40 per cent and the thirty-six-month figure rises by two-fifths, to about ₹3,300; retention is the strongest lever on the page. Then cut the tail churn to 1 per cent and see how little it moves the thirty-six-month figure compared with the formula tile, which leaps. That gap is why the formula flatters: it is most sensitive to the number you know least.
How long a lifetime you may claim
The a16z authors give a conservative rule: with only a few months of data, measure lifetime value as the value delivered to date, and prefer twelve- and twenty-four-month lifetime values to open-ended ones. A practical house rule follows from it. Claim no horizon longer than three times the months of data you have, and never more than thirty-six months before the company has thirty-six months of cohorts. Report the twelve-month figure next to whatever horizon you claim, because the twelve-month figure is almost entirely observed.
Use a discount rate that reflects what capital costs a company of your stage, not a bank rate. Fifteen per cent a year is a modest choice for a funded startup; a company borrowing against its receivables knows its own rate. And compute lifetime value by channel and by plan as well as for the whole base, because a cohort average blends customers from channels that keep them with customers from channels that do not, as the [channel CAC lesson](/library/blended-cac-lies-channel-level-truth) shows for acquisition.
Annual plans need one adjustment. A customer who prepays a year does not churn month by month; they renew or leave at the anniversary. Build their curve in yearly steps, count the full year’s contribution in the month it is collected, and do not mix them into the monthly curve, where they make early retention look far better than it is.
Lifetime value is not a forecast of how long customers might stay. It is the sum of what they have been seen to leave, extended only as far as the evidence reaches.
Using LTV without being seduced by it
Bill Gurley’s 2012 essay The Dangerous Seduction of the Lifetime Value Formula lists the ways the number misleads even when it is computed carefully. It is at best a good guess about the future. Its inputs move together: raise price and churn rises, spend more to grow and CAC rises. Customers bought with marketing tend to stay less well than customers who came on their own. And the promise that one day the company can stop spending and be remarkably profitable rarely comes true.
So use LTV for two narrow jobs. As a ceiling on CAC, with the twelve-month figure as the hard limit and the thirty-six-month figure as the stretch. And as a scoreboard for retention work, because a rise in the floor shows up in lifetime value faster than anywhere else. Do not use it to justify a CAC that the twelve-month figure does not cover unless the [payback lesson](/library/payback-period-and-cash-trap-of-fast-growth) arithmetic says your cash can carry the wait.
The quarterly LTV restatement
In the first week of each quarter rebuild the survival curve from every cohort at least three months old. Note the month-two figure, the floor and the tail churn of the floor, and compare each with last quarter. Recompute twelve-month and thirty-six-month lifetime value by channel and by plan using current contribution per customer. Write both beside the matching CAC.
Then replace the number in the deck, the board pack and the financial model with the restated one, and keep a dated line of every past figure so anyone can see it move. If the thirty-six-month figure fell, find which of the three inputs moved and give it an owner. If it rose because the floor rose, that is the retention work paying, and the CAC ceiling may rise with it, but only by what the twelve-month figure supports.
Nothing here is legal, tax or investment advice. The Pune app is illustrative and the survival model a simplification; your own cohort grid is the source of truth.
Sources
- Jeff Jordan, Anu Hariharan, Frank Chen and Preethi Kasireddy, 16 Startup Metrics, Andreessen Horowitz, August 2015 — LTV as present value of future net profit; the common mistake of using revenue; prefer 12- and 24-month LTV.
- Bill Gurley, The Dangerous Seduction of the Lifetime Value (LTV) Formula, Above the Crowd, September 2012
- Peter S. Fader and Bruce G. S. Hardie, How to Project Customer Retention, May 2006 — Rising retention rates as a sorting effect in a heterogeneous population; the shifted-beta-geometric model.