Analyzing Purchase Decisions Using Dynamic Location Data

Retailers’ efforts to monetize consumer location data are still dominated by naïve protocols such as geo-fencing – customizing marketing interactions based solely on the current location of all app users. While past research has noted the inefficiency of such approaches due to ignoring the heterogeneity of app users at any given location, alternative methods to monetize longitudinal consumer mobility data are scant in the literature. In this paper, we extend the methods of Ghose, Li, and Liu (2019) from continuously-measured indoor mobility trajectories to low-granularity urban mobility data. We determine consumers’ daily mobility trajectories, and for each day, we define the distance of each store from consumers’ “anticipated trajectories” by averaging store distance over all observed trajectories in the prior few days. To test our method, we model gas station choice during a six-month period in Staten Island, NY. Our data, also used to infer gas station visits, contain 11.7 million location records on 273 thousand devices observed near selected retailers including gas stations. We supplement our data with station-level daily fuel prices and, controlling for brand intercepts and station loyalty, estimate a conditional logit model to assess how consumers trade off gas prices versus store distance. We find that accounting for consumers’ location histories using our method allows for better out-of-sample predictive accuracy despite the generally strong station loyalty in the market. We also conduct counterfactual simulations, which reveal how the busiest traffic flows affect the geographic structure of competition in gasoline retail