Can Customer LTV Be Estimated Without Long-term Experiments?

The article examines whether customer lifetime value (LTV) can be accurately estimated without conducting prolonged experiments, focusing on the impact of delivery delays in food‑service platforms. Researchers from Stanford Graduate School of Business and Uber developed a model that links average order delay times to changes in repeat purchase probability, churn rates, and per‑order revenue. Using historical data from Uber Eats covering 1.2 million orders across multiple U.S. cities, the study estimates that a one‑minute increase in delivery time reduces LTV by approximately 0.5 % for the average customer, while a five‑minute delay can cut LTV by up to 3 %. The analysis also shows that the effect is more pronounced for high‑frequency users, whose LTV drops by 1.8 % per minute of delay.

Methodologically, the authors employ a causal inference framework that adjusts for confounding variables such as order value, time of day, and customer demographics. They validate the model against a hold‑out sample and compare predictions to results from a month‑long randomized controlled trial, finding less than 4 % deviation in LTV estimates. The study covers the United States from January to December 2024, encompassing both urban and suburban markets.

Key conclusions highlight that short‑term experiments can be replaced by data‑driven predictive models, enabling companies to forecast LTV impacts of operational changes in near real time. The findings suggest that investing in faster delivery logistics can yield measurable gains in customer value, with a cost‑benefit threshold identified at roughly 1.5 % LTV loss per minute of delay. This research offers a practical framework for mobile commerce platforms to optimize service quality and customer retention without extensive trial periods.