Customer Lifetime Value: What the Number Is Actually Claiming

Lifetime value appears in acquisition decisions as though it were a measurement. It is a forecast, it contains several assumptions, and the assumptions are usually invisible to the person acting on the number. A related distinction between output and value is covered in this guide on productivity versus efficiency.

Knowing which assumptions are in there is the difference between a useful planning figure and a licence to overspend.

What every LTV calculation assumes

That past behaviour predicts future behaviour. The number is derived from customers who bought before. If the product, price, market or acquisition mix has changed, the historical pattern is describing a different population.

A retention or churn curve. Almost always the largest source of error. A small change in assumed retention compounds heavily over a multi-year horizon.

A time horizon. Three years, five years, forever with a discount rate. The choice is arbitrary and it dominates the answer.

A margin assumption. Revenue LTV and contribution LTV differ enormously, and the version that supports an acquisition decision is the second one — you cannot spend revenue on advertising.

A discount rate, if the horizon is long. Money in year four is not worth money today, and most working LTV figures quietly ignore this.

Uniformity across the cohort, when the average is used. Which is the problem below.

The average is the least useful form

Customer value is heavily skewed. A small share of customers accounts for a large share of value, which means the mean sits well above the median customer.

An average LTV of £180 is consistent with most customers being worth £40 and a few being worth thousands.

The consequence for acquisition: if you bid up to £60 to acquire an average customer, you are overpaying for most of the customers you actually get. The average is real and no individual acquisition is average.

Report the distribution. Median, the quartiles, and the share of value in the top decile. Three extra numbers, and they change how the figure is used.

Better still, report LTV by segment — acquisition channel, first product, geography, first order value. The variation between segments is usually larger than any campaign optimisation you were considering, and it is the actionable part.

Cohort curves beat point estimates

The most useful version of this metric is not a number at all.

Plot cumulative revenue per customer against months since acquisition, one line per acquisition cohort.

This shows:

How fast value accrues, which determines how long you can afford to wait for payback.

Whether recent cohorts differ from older ones. If the last three months' cohorts are tracking below previous ones at the same age, acquisition quality is declining — and this is visible months before it shows in an aggregate LTV figure, which averages over years of history.

Where the curve flattens, which tells you what horizon is meaningful. If nothing accrues after month eighteen, a five-year LTV is mostly extrapolation.

This chart is more useful than any single LTV number, and it is straightforward to produce from an order table.

The payback period is the operational number

For deciding acquisition spend, LTV is frequently the wrong metric and payback period is the right one.

How many months until a cohort's cumulative contribution covers its acquisition cost?

Why it is better for decisions: it depends on the near part of the curve, which is observed rather than forecast. A three-year LTV depends on retention assumptions for years two and three; a six-month payback depends on six months of actual data.

It also constrains growth realistically. A business with a twenty-month payback needs to fund twenty months of working capital per customer, regardless of how attractive the LTV looks.

Where LTV gets misused

As a licence to bid up. "LTV is £200 so we can pay £100 to acquire" ignores which customers the channel actually brings, whether the LTV was contribution or revenue, and how long the money takes to arrive.

Applied uniformly across channels. Customers acquired from different channels have different retention. Applying the blended LTV to every channel systematically overpays for the channels bringing worse customers, which is usually the channels that look cheapest.

Extrapolated beyond the data. A five-year LTV computed from eighteen months of history is three and a half years of assumption. This is extremely common in businesses that have not existed for five years.

Confused with incrementality. A high-LTV cohort from a retargeting campaign may consist of customers who would have bought anyway. LTV tells you what a customer is worth; it says nothing about whether the channel caused them to become a customer. See incrementality.

Never revised. A figure calculated two years ago, still in the acquisition model, describing a business that has changed.

Making the number honest

Use contribution, not revenue. After cost of goods, payment fees, shipping, returns and support.

Subtract returns properly. In categories with high return rates, gross orders overstate value substantially.

State the horizon and show how sensitive the number is to it. If three-year and five-year LTV differ by 60%, the horizon is doing more work than the data.

State the retention assumption, and test the number against a pessimistic one. If the decision reverses under a slightly worse curve, the decision is not supported.

Recalculate quarterly, and compare predicted against actual for cohorts old enough to check. This is the only real validation there is, and almost nobody does it.

Segment by acquisition channel, at minimum.

What to present alongside it

Given all the above, an LTV figure alone is not a defensible artefact. What travels with it:

The cohort curve, so the shape is visible.

The median and the distribution, so nobody treats the mean as typical.

The payback period, which is what the finance conversation actually needs.

The horizon and the retention assumption, stated.

And by segment, because the between-segment variation is the useful information.

The summary

LTV is a forecast with a retention assumption, a horizon and a margin definition inside it. Whoever built it made those choices and whoever uses it usually does not know what they were.

The average describes almost nobody, because the distribution is skewed. Report the median and the segments.

The cohort curve is more useful than the number, and it shows quality declining months earlier.

Payback period is the better operational metric, because it depends on data you have rather than on assumptions about years you have not observed.

And LTV says nothing about whether a channel caused the acquisition — which is the question the acquisition budget actually turns on. For an implementation example, see the Google Analytics user-lifetime documentation.