Revenue Analytics
Part of Ecommerce customer value analysis
Reporting uncertainty in lifetime-value estimates
Separate observed customer value from forecasts, label scenario ranges correctly and assess estimates against later outcomes.
A lifetime-value estimate is a forecast: show its horizon, value measure and uncertainty beside observed value. Never present a projected A$ amount as sales that have occurred.
Separate recorded and future value
Report three figures under compatible customer and amount rules:
- Observed value:qualifying value recorded through a stated customer age and adjustment cut-off.
- Projected additional value:estimated value for a stated future period, with its data cut-off and assumptions.
- Projected total:observed value plus projected additional value.
If the decision concerns contribution, a forecast of sales alone is insufficient: the relevant future costs need their own estimate and coverage statement.
If a customer report includes expected purchase value, label it as a forecast and state its horizon, measure and uncertainty; do not present it as recorded value.
Observed vs. Projected Customer Value: Key Components in Lifetime Value Reporting
- Observed Value
- Recorded value from customer activity up to a stated cut-off date
- Projected Additional Value
- Estimated future value over a defined horizon, with assumptions and data cut-off
- Projected Total Value
- Observed Value + Projected Additional Value
Label the uncertainty honestly
State the forecast horizon, cohort size, age when forecast, method version, training period and excluded records. If the method supports a lower and upper outcome, say what that range means. A range built from chosen business assumptions is a scenario range.
A statistical prediction interval needs a method for uncertainty about the specified future outcome and a stated coverage level. A confidence interval for an estimated historical mean addresses a different target.
For illustration, an observed amount per entrant over a stated period plus additional value under two documented purchase scenarios yields a projected total range. This is a scenario range, not a statistical prediction interval or a store result.
If no defensible range is available, say that the range was not estimated and avoid treating the point forecast as precise.
Elements to Include When Reporting Lifetime Value Uncertainty
- Cohort Size
- Number of customers in the forecast group
- Training Period
- Time period used to build the forecast model
- Excluded Records
- Reasons for excluding certain customer data (e.g., refunds, churned users)
Compare forecasts with later outcomes
Before a material decision, a proposed check is to forecast earlier customer groups at the same age as today's group, then compare those forecasts with their subsequently recorded value over the same horizon. Report the groups checked and the size and pattern of the errors. Do not describe a backtest as completed without results.
Changes in product mix, prices, offers, customer identity coverage or refunds can make old data less representative. A narrow model interval does not automatically cover those business changes.
If plausible outcomes lead to different decisions, limit the commitment or collect more follow-up. Preserve the forecast version and later observed result so revisions can be explained.
Backtesting Forecast Accuracy: Steps to Validate Lifetime Value Estimates
- Step 2Apply Same Forecast Model — Use identical method, assumptions and training period as current forecast
- Step 4Report Error Patterns — Document size and direction of errors across cohorts
- Step 5Review Model Relevance — Assess if changes in pricing, offers or product mix affect model representativeness
