Improving Precision in Cluster-Level Pricing Experiments with Temporal Switchbacks

Online pricing systems often cannot randomize treatments at the user level because prices are deployed over operational units such as routes, regions, or markets. This creates low-powered A/B tests due to persistent cluster heterogeneity, seasonality, and autocorrelated demand. We study temporal switchback designs for cluster-level pricing experiments, where each cluster alternates between treatment and control over time. Using LATAM Airlines ancillary-pricing data and calibrated simulations, we show that switchbacks can substantially reduce standard errors relative to parallel cluster designs. We further characterize how switching cadence interacts with temporal dependence, highlighting when daily or weekly switchbacks are preferable. The results provide practical guidance for experimentation in pricing and transportation platforms where fine-grained randomization is infeasible.