How to Calculate Customer Lifetime Value (LTV): Formulas and Examples

A.The formulas, and the term that does the work

The general form is average revenue per customer per period, multiplied by the number of periods the customer stays. E-commerce usually decomposes the first term into average order value times purchase frequency, and both versions have a profit-based variant that multiplies by gross margin. Subscription businesses invert the lifespan term instead of estimating it directly, because a constant monthly churn rate implies an average lifespan of one over that rate. So LTV becomes average monthly revenue per account divided by monthly churn, or annual contract value divided by annual churn.

Notice what is observed and what is not. Average revenue per customer is a fact you can read off last quarter’s billing. Gross margin is a fact your finance team already owns. Lifespan is not a measurement at all. It is a forecast, and every version of the LTV formula is a different way of dressing that forecast up as arithmetic.

The dressing is convincing because the churn form looks empirical. A 3% monthly churn rate is measured, so $100 divided by 0.03 feels measured too. It is not. Dividing by a churn rate asserts that the same 3% applies in month one and in month forty, which is the one thing customer retention curves reliably do not do.

B.Where the forecast breaks, in four predictable ways

Churn is not constant. It is highest in the first periods and flattens as the surviving population self-selects toward the committed. A single blended rate taken from a mixed-tenure customer base therefore understates the lifespan of anyone who makes it past the early cliff and overstates it for everyone who has just signed up. For a growing company, where new customers dominate the base, the blended rate is dragged toward the early-churn figure and LTV comes out too low. For a shrinking one it comes out too high. Our churn diagnosis guide covers reading the curve rather than the average.

Averages hide a skew. Customer value distributions have long right tails, so the median customer is worth well below the mean. If the top decile accounts for half of total value, the mean describes a customer who does not exist, and using it to justify a per-customer acquisition bid overpays for the typical one.

Revenue and profit get mixed. The most damaging version of this is not inside the LTV number at all; it is the inconsistency between LTV and the cost it gets compared against, which the next part takes up.

Future money is counted as present money. A customer generating $1,200 across one year and one generating $1,200 across five have identical nominal LTV and materially different value. Applying a discount rate matters most for long subscription relationships, and barely at all for a business whose customers are done inside a year.

There is an escape from all four, and it is not a better formula. Historical LTV, the actual revenue a customer has already generated, has no forecast in it and therefore none of these errors. Its limitation is that it only describes customers old enough to have a history, which is why predictive models exist: they use first-30-to-90-day behaviour, first product bought and acquisition channel to estimate value early enough to act on. Both are answers to the lifespan problem. Neither removes it. Part III is about the version that comes closest.

A.LTV:CAC, and the inconsistency that inflates it

The ratio of lifetime value to customer acquisition cost is what tells you whether the model works. Below 1:1 you are paying more for customers than they will ever return. Between 1:1 and 2:1 there is no room for operating costs or for being wrong. Around 3:1 is the widely cited healthy band. Above 5:1 the usual reading is not excellence but underinvestment: you could afford to buy more customers than you are buying.

The band is easy to hit accidentally, because the two sides of the ratio are usually prepared by different people. Marketing supplies a CAC, which is a real cost. Finance or analytics supplies an LTV, which is frequently a revenue figure. Divide one by the other and the ratio is not wrong by a rounding error, it is wrong by your gross margin.

Pick gross profit, and hold it constant everywhere the ratio appears. A business reporting 3.3:1 on revenue and operating at 1.3:1 on profit is not slightly optimistic, it is spending as though it had two and a half times the headroom it has.

B.Blended is a health check, by channel is an instruction

Even a correctly constructed blended ratio only tells you whether the company is solvent. It does not tell anyone what to do on Monday, because nobody buys blended traffic. The version that changes behaviour takes the average LTV of customers acquired through each channel and divides by the cost to acquire through it.

Expect the spread to be wide enough to reverse decisions: organic search at 5:1 while paid social sits at 1.5:1 is an ordinary result, and it is the opposite of what first-purchase ROI reports. Channels that bring in loyal repeat customers look expensive on day one and cheap on day four hundred, which is precisely the horizon a first-purchase ROI calculation cannot see. This is also where the segmentation argument bites hardest. A blended $1,000 LTV covering enterprise at $5,000 and SMB at $300 supports no bid on either. Tracking cost per acquisition by channel is the other half of the same exercise.

The prerequisite is a join that most stacks do not have by default: the acquisition source of a person, held from their first anonymous visit, still attached to them when they buy for the fourth time eleven months later. Session-scoped analytics loses that link at the first device change. Person-level tracking exists to keep it.

A.Read down the column, not across the row

Cohort LTV groups customers by acquisition period and tracks cumulative revenue or gross profit per customer at fixed ages: 30, 60, 90, 180, 365 days. January’s cohort has its own curve, February’s has another, and the comparison that matters is between the same age in different cohorts rather than between different ages in one.

Two things fall out of that grid that no average produces. The first is direction: if each successive cohort recovers faster at 90 days, onboarding or retention or pricing is improving, and if it recovers slower, something is degrading while the blended figure still looks fine. The second is the payback date. Cross the row against your CAC and you get the month the customer pays for themselves, which is the number that governs how aggressively you can spend, because it is a cash flow constraint rather than a profitability one.

The grid also disposes of the lifespan problem from Part I, in the only honest way available. It does not forecast a lifespan. It reports observed cumulative value at an age you have actually reached, and leaves the cells you have not reached empty. For the full technique see our cohort analysis guide and the e-commerce version.

B.What the curve tells you to change

The shape of the curve names the lever, which is more than a scalar LTV can do. Three levers exist, and they show up in different places on the grid.

A low first cell is an order value problem. Cross-sell, bundles and tiered pricing raise it, constrained by relevance: a recommendation that is a revenue grab damages the later cells to lift the first.

A curve that rises slowly is a frequency problem. The specific milestone worth instrumenting is the second purchase, which is widely reported to predict a third far more strongly than the first predicts a second, and it is the cheapest cohort intervention available: post-purchase sequences, replenishment prompts, and for subscription products, depth of usage rather than more logins.

A curve that flattens early is a lifespan problem, and it is the only one of the three that compounds. Onboarding that reaches value quickly moves the whole curve up permanently for every cohort behind it. Behavioural churn signals let you act before the flattening is visible. Populations is the mechanism for holding “customers matching the declining pattern” as a reusable definition, so the same group can be read in Cohorts, in Revenue, and in Campaign Performance rather than rebuilt three times with three slightly different filters. Our retention strategies guide works through the interventions.

Bain’s finding that a 5% improvement in retention raises profits by 25% to 95% is usually quoted at the start of articles like this one. It belongs at the end, because it is a statement about the third lever specifically, and you cannot tell whether it applies to you until the cohort grid shows you where your curve actually flattens.