Ignoring consideration sets in modeling customer purchase decisions may lead to biased estimation of customer preferences, yet consideration sets are difficult to infer in brick-and-mortar contexts. We show that the challenge of estimating consideration set models in brick-and-mortar contexts can partially be overcome with an emerging source of data: “heatmap data” collected using in-store sensors. In contrast with clickstream data in e-commerce settings, which identify and track individuals, heatmap data show customer traffic only in the aggregate. Despite this limitation, we show that heatmap data enable us to recover many of the benefits of individual-level data. In a setting in which consideration set probabilities are non-parametric, heatmap data enable identification of consideration set models that cannot be estimated using aggregate sales data alone. In a setting in which consideration set probabilities are parameterized, we show that heatmap data enable us to estimate a consideration set model that captures the effects of any variable on both consideration and choice decisions without imposing exclusion-restriction assumptions on customer behavior. We also demonstrate that heatmap data can lead to decreased finite-sample bias. Finally, we install heatmap sensors in an apparel retail store and find that heatmap data can result in distinct coefficient estimates, improved predictive accuracy, and better estimated revenues in a product placement decision problem.
Note: Research papers posted on SSRN, including any findings, may differ from the final version chosen for publication in academic journals.