Customer LTV:
how to calculate lifetime value

What is LTV, and which value should you measure?

LTV, or Lifetime Value, is a customer's value over their relationship with a business. It can be measured as revenue, gross profit or contribution toward costs. A statement such as “LTV is $200” is therefore of little use without a calculation definition. The report should identify what the amount includes, the customer group, the observation period, and which money has already been earned versus what is still expected.

Imagine two equipment buyers. One places a large order and never returns. The other begins with a small purchase, then buys consumables and servicing. The first order helps assess the initial sale; LTV shows how the relationship develops. Repeat revenue alone does not prove profitability: orders also have product, delivery and other costs.

For the marketing decisions below, we use contribution before acquisition costs: sales less product cost, variable fulfilment costs and retention costs. We compare CAC with that result separately. This is a management method, not a substitute for financial accounts.

Separate observed value from future forecasts

A business that has observed customers for six months knows their accumulated value over six months. That is not automatically their lifetime value. An accurate label is “180-day contribution LTV, observed”. You can show a one-year forecast alongside it, but in a separate column with the calculation date and assumptions.

Stores can use windows of 30, 90, 180 and 365 days after the first purchase. Subscriptions can use a number of billing periods; B2B can use a horizon aligned with renewals or repeat orders. Choose the horizon to match the business cycle, not because a longer period produces a more attractive number.

A new cohort has not yet lived through all future periods. You cannot compare its 30-day accumulated result with a full year for an older group. Nor should you multiply month one by twelve without validation: first purchases, repeat orders and seasonal renewals occur at different rates.

What counts as a customer's “lifetime”?

A subscription has a cancellation event, although the customer may return. In retail, the latest order does not necessarily mark departure: someone may buy once a year. Define an inactivity rule separately, test it against history and do not treat it as a known end date for the relationship. With limited history, use an observed horizon and several continuation scenarios.

Bring each customer's purchases together

Your minimum export should let you trace each transaction back to its customer. Include a stable customer ID, order ID, first-purchase date, transaction date, amount after discounts, refunds, cost of sales, variable expenses and acquisition source. For B2B, decide whether a customer is a person, company or corporate group. Use the same definition for LTV and your CAC calculation.

1. Combine sales channels. Purchases in a store, app, website or through a salesperson should share one history when you can reliably identify the customer. Do not merge different people merely because their names are similar.
2. Remove technical duplicates. An order-status update or repeated import must not create another sale. Payments and orders are different entities: several payments can belong to one purchase.
3. Account for cancellations and refunds. Allocate them back to the original cohort. Use one rule for the period in which adjustments are recorded and state the report's data cutoff date.
4. Check that the history is complete. The first purchase recorded after a CRM migration may be a person's tenth purchase. If the history is incomplete, call the date “first known purchase” and limit conclusions to the available data.

Internal IDs are sufficient in the calculation spreadsheet; names, phone numbers and addresses are unnecessary. To identify sources, use a consistent UTM tagging process, and manage relationships in a CRM with an agreed data model.

If sales data is sent to Yandex Metrica, start with the official instructions for uploading CRM data and the “Customers” report. These tools help connect customer and order records. You still need to define the LTV method, cost of sales and cohort rules separately; having a report does not mean you have a ready-made profit estimate.

Calculate LTV without unnecessary assumptions

A cohort's observed result

A cohort is a fixed group of customers, such as everyone who first bought in January. Add the chosen financial result of their transactions over the same time after each first purchase. Divide by the original number of customers in the group, including those who never returned.

An average-order-value breakdown

For the same orders and customers, you can express the result as average order value × orders per customer × contribution as a share of revenue. Average order value is revenue / orders; orders per customer is orders / original customers; the contribution share is total contribution / revenue. With these consistent definitions, the product equals contribution per customer. Calculate the share from totals, not by taking an unweighted average of product margin percentages.

Another approach is order value × purchase frequency × relationship duration. This requires matching units: purchases per month are multiplied by months. If duration is an assumption, the result is a forecast. With mixed products and changing costs, it is more reliable to calculate each order's result first, then average across customers.

A general overview of LTV is available from Stripe. When comparing formulas from different sources, always check the definitions: revenue, gross profit and contribution give different answers even for the same customer base.

Example: 100 customers observed for six months

A teaching example for a components store. The cohort contains 100 new buyers, with transactions counted over each buyer's first 180 days. All 100 have completed that observation window. Sales are shown after discounts and refunds; the same tax and cost conventions apply to every row.

In Excel or Google Sheets, enter B2–B7 as numbers and apply a currency cell format where appropriate. In B8, enter =B4-B5-B6-B7, in B9 enter =B8/B2, and in B10 enter =B4/B2. A cohort size of zero means the metric cannot be calculated, not that customer value is zero.

Cross-check with a second method: average order value is $12,000 / 240 = $50; orders per customer are 240 / 100 = 2.4; contribution share is $3,800 / $12,000. Multiplying $50 × 2.4 × that share again gives $38. Use the original figures without rounding intermediate steps.

If acquiring these 100 buyers cost $2,000, CAC is $20. After acquisition, $1,800 remains for the whole cohort, or $18 per customer. This is not net profit: the example has not yet allocated fixed costs.

Compare channels at the same customer age

First group buyers by the agreed acquisition source, then by first-purchase date within each source. Fix a rule such as first known acquisition or your chosen attribution model. Do not relabel an existing customer's source as remarketing just because a repeat order followed an ad.

A has the higher final contribution after acquisition; B recovers acquisition costs sooner at the observed checkpoints. Budget allocation also depends on cash reserves, available demand, data quality and how costs change as volume grows. These figures do not prove that channel A can absorb unlimited extra budget.

In a cohort table, a blank cell means “this period has not yet been observed”; zero means “the period is complete, with no result”. These are different states. Compare fully observed windows or apply an explicitly documented forecasting method. For small groups, show customer counts and the influence of large orders separately: one deal can shift the average substantially.

When the churn-based formula is useful

For a stable subscription business, one common estimate is Revenue LTV ≈ monthly revenue per subscriber / monthly subscriber churn rate. A version of this approach is used in Stripe Billing's LTV calculation. This is an estimate under specific assumptions, not cash already received. Stripe Billing uses an assumed 60-month lifetime when its measured churn is zero.

Teaching example: $18 per month per paying subscriber, a 70% contribution share after variable costs, and 5% monthly churn. With constant conditions, the simplified contribution estimate is $18 × 0.70 / 0.05 = $252. Here, 5% is entered as 0.05 and refers to customers leaving, not a percentage decline in revenue.

The formula assumes a constant probability of leaving and comparable contribution across periods. It ignores discounting and does not model complex plan changes. It is particularly unreliable when new users leave more often than established ones, or when there are seasonal pauses, annual billing, refunds or frequent reactivations. At zero observed churn, the division itself is undefined; zero churn over a short window does not imply an infinite customer lifetime.

For a basic churn calculation, count customers who left from the group active at the start of the period and divide by that group's size. Do not mix new sign-ups into the original base. A definition and example are published by Stripe. If conditions are unstable, chart actual accumulated contribution and forecast its continuation separately.

Connect LTV, CAC and the payback period

In the main example, LTV/CAC = $38 / $20 = 1.9. The numerator is 180-day contribution before acquisition; the denominator is the fully loaded acquisition cost per customer in the same cohort. Dividing $120 of revenue by the same CAC gives 6, but that is a different measure that hides order-fulfilment costs.

Do not set an affordable CAC solely by a “must be 3:1” rule. Businesses differ in purchase frequency, costs, risks, working-capital needs and forecast maturity. To set a useful limit, decide how much contribution should remain after acquisition and how quickly those costs must be recovered.

For example, with an observed 180-day LTV of $38, a business may want to retain $15 for fixed costs, its planned return and a reserve. Under that model, its CAC limit would be $23. This is a management decision in a teaching example; the buffer and time horizon for your own business need their own justification.

Find when acquisition costs are recovered

In the channel table, A crosses CAC between day 30 and day 90, but those three observations do not reveal the exact date. You need a more detailed cumulative curve. Even that crossing means acquisition payback within the chosen contribution model, not payback for the entire business. With payment delays, also check cash flow: recorded profit and money in the bank may arise on different dates.

The performance marketing this helps align budget limits with actual sales. A high forecast LTV does not remove the need to monitor actual CAC and the ability to pay current costs.

What to do when your history is incomplete

Start with observed results over the available 30 or 90 days. Build conservative, base and more optimistic repeat-purchase scenarios. State where the assumptions come from: older comparable cohorts, renewal agreements or a limited pilot. Do not present your chosen scenario as guaranteed income.

To test the model, return to a historical date. Use only data that was known then, forecast the following months, and compare the prediction with purchases that subsequently occurred. If the model accidentally sees future orders, the accuracy assessment is meaningless. Examine errors by channel and segment, not just in the total.

For longer horizons, consider separately whether future financial results should be discounted to present value. The discount rate and timing must match the model's periods. Do not insert an arbitrary percentage to make the formula look sophisticated: first establish complete orders, correct costs and a reliable behavioural forecast.

Automation and AI analytics can help refresh calculations and identify patterns. But an algorithm cannot fix unknown product costs, duplicate customers or incomplete history without data sources and clear rules.

Increase LTV and test whether changes help

Address why customers do not return or generate little contribution. First check whether you deliver on your promises: product availability, lead times, ease of use and issue resolution. Then consider timely reorder reminders, a suitable bundle or ongoing servicing. The offer should fit the customer's needs and the agreed communication preferences.

More discounted purchases can coincide with lower contribution. An extra service can increase order value while overloading support. Define the main metric, time horizon and guardrails in advance: refunds, complaints, servicing costs and communication opt-outs. Evaluate all customers assigned to the change, not only those who responded.

A teaching test of a repeat-purchase programme

Two comparable groups of 500 customers generated observed 90-day contribution of $26 and $28 per original customer. The second group's programme cost an extra $0.80 per person, not yet included in the metric. Adjusted contribution is $27.20; the difference from the first group is $1.20, or $600 across 500 customers. If those costs have already been deducted, do not deduct them a second time.

A difference in averages alone does not prove that the programme caused the change. You need sound allocation, enough observations and an assessment of uncertainty. Experiment planning is covered in our A/B testing guide. If you compare different months without an experiment, account for seasonality, prices, product range and acquisition channels, and describe the conclusion as an observation rather than a proven cause.

Checklist before using LTV in a budget

Add a data owner, cohort sizes, export source and a log of methodology changes to the regular report. If LTV falls after duplicates are corrected, the business may not have deteriorated: recalculate comparable historical periods first. A useful metric supports decisions rather than rising because its definition changed.

When data is scattered across departments, start with a CRM implementation plan: agree on events, owners and checks that sales data is arriving. Then automate the final report.