Issues with shopper tracking and data quality: From solving multi-floor issues to driving better positional accuracy

One of the most popular methods for tracking the in-store shopper journey is through the electronic detection of smartphones. While there are various technologies for doing this, most share the same basic approach: using multiple sensors to monitor for phone signals. By analysing the signals received at each sensor, it is possible to backtrack and establish the location of the phone. Unfortunately, this process has numerous challenges in most real-world settings. Physical locations are difficult environments for electronic measurement. Floors add a three-dimensional problem to shopper positioning that breaks many systems. Blockages, reflections and even the phone’s location on the customer can influence signal readings by individual sensors. These factors make positioning based on sensor readings unreliable and erratic. Store and associate devices further complicate the picture. Unless these devices can be identified and filtered, they create a false picture of actual shopper movement. None of these problems is easy to solve with traditional processing techniques. However, they are problems for which machine learning is highly suitable. This paper will describe each problem, explain its significance and then outline a machine-learning approach for solving it.