Hamidreza Badri and Allen Tran
Suppose a subscription business persuades a customer not to cancel. How much was that intervention worth? The usual answer is the customer’s remaining lifetime value: all the revenue they are expected to generate from this point onward.
That answer is intuitive, useful—and generally too large.
The problem is the counterfactual. A person who cancels does not vanish from the economic universe. They may return next month, choose a different plan, or rejoin after a year. Even someone who has never subscribed may do so later without the intervention. Conventional lifetime value counts the future of the subscriber but assigns a value of zero to this alternative future.
What a business actually creates through acquisition or retention is the difference between two trajectories: expected revenue when the person is on the service and expected revenue when they are off it. We call this incremental lifetime value. It is the same causal instinct used in an experiment—compare what happened with what would otherwise have happened—applied to a customer’s stream of future states.
We model those states with a finite Markov chain. A household can be a new subscriber, a long-tenured subscriber, a former subscriber, or a customer on one of several plans. Estimated transition probabilities describe renewal, cancellation, rejoining, and switching. Chaining the transitions together recovers expected revenue from both on-service and off-service states, even when information about non-subscribers is limited.
This reframing is more than an accounting correction. Once subscriber behavior is represented as a state-transition system, the same model can produce long-range subscriber forecasts that respect the size of the market, estimate the revenue consequences of price changes, and compare policies such as targeted discounts.
Lifetime value asks, “What is this customer’s future worth?” Incremental lifetime value asks the more useful question: “How much of that future did our action actually create?” For investment decisions, the difference between those questions can be the difference between growth and expensive self-congratulation.