How to calculate customer lifetime value: 4 methods compared
In short
Customer lifetime value is average order value × orders per year × lifespan × margin. Only lifespan is a guess, and methods fill it differently: on identical store data, four common methods give a CLV from 24 to 70. For deciding what to pay for a customer, use historic CLV on contribution margin over a horizon you have observed.
The CLV formula for ecommerce, and which inputs are guesses
The formula on almost every page is order value × purchase frequency × lifespan, multiplied by margin to turn revenue into profit. Three of those four inputs are measured. The fourth, lifespan, is a forecast wearing the costume of a measurement.
| Input | Where it comes from | Measured or assumed? |
|---|---|---|
| Average order value | Order history | Measured |
| Orders per year per active customer | Order history | Measured |
| Margin | Your cost structure | Measured |
| Lifespan | How long customers stay active | Assumed, every time |
Lifespan cannot be observed for customers who are still buying, and in a young store that is most of them. Any method that produces a CLV has to decide how to fill that gap, and the four common methods decide differently. That is why they disagree.
Why CLV estimates differ so much: four methods, one store
Take one store and hold every input fixed:
| Input | Value |
|---|---|
| Average order value | 60 |
| Orders per year, active customer | 2 |
| Contribution margin | 35% |
| Annual margin per active customer | 60 × 2 × 0.35 = 42 |
| Retention, year to year | 40% |
| Orders in the first 12 months, per new customer | 1.5 |
| Discount rate | 10% |
Now calculate CLV four ways, all on those numbers:
| Method | Formula | CLV |
|---|---|---|
| Simple, lifespan = 1 ÷ churn | 42 × (1 ÷ 0.6) | 70.00 |
| Discounted, first year included | 42 × 1.1 ÷ (1.1 − 0.4) | 66.00 |
| Historic, first 12 months only | 1.5 × 60 × 0.35 | 31.50 |
| Discounted, as commonly printed | 42 × 0.4 ÷ (1.1 − 0.4) | 24.00 |
The same store, with the same customers and the same data, produces a CLV anywhere from 24.00 to 70.00, a spread of nearly three times.
Nobody made an arithmetic error. The bottom row is the formula Shopify's guide prints, margin × retention ÷ (1 + discount − retention), and it is correct for what it measures: expected margin from *future* years only. It leaves out the first year, the one that has already happened, which is why it sits so far below the discounted row above it. Neither the guide nor any page on the first screen says so, and a reader who copies it undervalues every customer by the entire first year.
Pass a CLV between teams and the method is lost on the way. By the time it reaches a budget meeting it is "our CLV is 70", and the person setting the acquisition target has no idea it could equally have been 24.
Historic vs predictive CLV
Historic CLV adds up what customers have actually generated over a fixed window. Predictive CLV extrapolates beyond what has been observed, using a retention assumption. The trade is certainty against completeness.
| Historic | Predictive | |
|---|---|---|
| Uses | Only orders that happened | A retention rate projected forward |
| Error | Understates, because customers are still buying | Over or understates, depending on the assumption |
| Moves when | Customers actually buy | You change an assumption |
| Safe for | Setting a floor on acquisition cost | Planning, clearly labelled as a forecast |
In the example above, the historic figure of 31.50 is the only one of the four you could defend in an audit. It is also the lowest defensible number, because it ignores every order after month twelve. That is a feature when the question is "how much can I safely pay for a customer?" and a flaw when the question is "what is this customer base worth?"
CLV time horizon: how many years to count
Count only as many years as you have watched customers for. A store with two years of order history has no evidence about year five, and a CLV built on year five is a guess about customers who do not exist yet.
The simple 1 ÷ churn method has no horizon at all: it assumes the current retention rate holds for ever. At 40% retention that is mostly harmless, because the tail shrinks fast. At high retention it is not, and the formula happily credits you for decades of orders from people who will have moved house, changed habits, or forgotten you.
A practical rule: cap the horizon at the age of your oldest reliable cohort, and state the horizon next to the number every time.
The input that moves CLV most
Retention. Hold everything else in the example constant and move only the retention rate by ten points in either direction:
| Retention | Simple CLV |
|---|---|
| 30% | 42 ÷ 0.7 = 60.00 |
| 40% | 42 ÷ 0.6 = 70.00 |
| 50% | 42 ÷ 0.5 = 84.00 |
A ten-point swing in one input moves the answer from 60.00 to 84.00. Order value and frequency are known almost exactly; retention is estimated from the most recent cohorts, which are the smallest and noisiest you have. The number everyone quotes is therefore most sensitive to the input measured worst. Measuring retention by cohort rather than as one blended rate is what narrows it.
Which CLV to use when setting what you pay for a customer
Use the historic figure over a horizon you have actually observed, on contribution margin rather than revenue. This is the whole of unit economics in one decision: what a customer is worth sets what you may pay for one. It is the lowest of the defensible numbers, which is the point: an acquisition target built on it cannot be undone by a retention assumption that turns out wrong.
The difference shows up against a widely repeated rule of thumb that CLV should be three times acquisition cost. At an acquisition cost of 20, the simple method gives a ratio of 3.5 and says you are safe. The printed discounted formula gives 1.2 and says you are in trouble. The historic figure gives about 1.6. Same store, three verdicts.
The better question is not the ratio but the time: how many months until a customer's margin repays what they cost. That is the CAC payback period, and unlike CLV it does not need a forecast to answer.
Common questions
Should CLV use revenue or margin?
Margin, and specifically contribution margin. A revenue CLV compared with acquisition cost ignores the cost of the goods and delivery that the revenue paid for, and makes every customer look several times more valuable than they are.
How often should I recalculate CLV?
Quarterly is enough for most stores. Recalculate sooner after a price change, a shift in acquisition channel, or a new product line, because each changes who your customers are and therefore how they retain.
Is predictive CLV from a tool more accurate?
It can be better at spotting which customers will return, but it still rests on a retention assumption learned from your recent cohorts. Ask the tool for its horizon and its method before comparing its figure with anyone else's.
What is a good CLV for an ecommerce store?
There is no universal figure, because CLV only means something against what a customer costs to acquire. Compare it with your own acquisition cost and payback period rather than with another store's number.
See it on your data
On a twenty-minute call we connect one ad account and one month of your orders, and work out the gap between what the platform claims and what your shop took. Profit and POAS stay blank until you enter your margin and costs per order.
Book a demo