Data & Analytics Practical insights
Separate predicted customer value from campaign lift
Use customer lifetime value predictions carefully: distinguish likely future value from the change a marketing action can cause, then test the decision.
A customer lifetime value model can correctly identify valuable customers and still recommend an inefficient campaign audience. Predicting who will spend is different from predicting whose spending a particular action can change. Keep both questions explicit when AI decisioning turns customer scores into marketing choices.
Define a useful value target
Choose a future horizon and a commercial measure, preferably one that accounts for the costs relevant to the decision. Exclude information that would not be available when the prediction is made. Validate forecasts on later cohorts, and check performance across customer groups rather than relying only on a strong overall ranking.
Recognize the intervention gap
Imagine two customers with similar predicted future margin. One is likely to reorder without encouragement; the other may respond to a helpful reminder. A discount offered to both could reduce the first customer’s margin without changing behavior. This is an illustrative mechanism, not a claim about any real audience. High predicted value alone does not reveal the incremental effect of contact.
Test the targeting policy
Compare the proposed value-based policy with the current approach using an appropriate randomized design. Evaluate outcomes across everyone assigned, including contact costs, incentives and potential customer fatigue. When sufficient suitable data and expertise exist, treatment-effect methods can investigate how response differs across customers. EconML documents approaches that explicitly model treatment contrasts rather than ordinary outcome prediction. [1] Those methods still require defensible assumptions and validation.
Use each model for its proper purpose
Retain CLV forecasts for planning, service prioritization or reporting where justified; predictive usefulness does not disappear because it is not causal. For marketing allocation, require evidence that the resulting decisions improve future value. Document the decision rule, its test and the conditions where it should not apply. Start by comparing one proposed audience policy with the business alternative before deploying value scores throughout every customer journey.
Sources and evidence
Sources checked on 4 October 2026. Proposed workflows and hypothetical examples are editorial analysis.
From insight to practice