Using Wi-Fi probe requests from mobile phones to quantify the impact of pedestrian flows on retail turnover

This paper discusses the opportunities afforded by novel population sensing technologies in the field of 'smart' urban management. In particular, it focuses on the application of these new sources of data in retail analysis. Our goal is to integrate data derived through novel pedestrian counting and point-of-sale systems to build a statistical model that captures the relationship between retail turnover and footfall in the UK. The point-of-sales data are provided by two UK-based food & beverage retailers. To accurately measure the pedestrian activity around retail units, we make use of the data generated by the SmartStreetSensor project: a deployment of a large network of sensors installed across 105 towns and cities in the UK that collect Wi-Fi probe requests generated by mobile devices. We propose and implement novel methods for processing these raw signals into accurate estimates of pedestrian activity without compromising participants' privacy. The resulting data is then integrated into seasonal ARIMA and dynamic regression models that can be used to predict future sales. Our results indicate that the dynamic regression model that accounts for fluctuations in footfall data outperforms seasonal ARIMA model that uses only past values and behaviours of transaction data to predict future sales. Thus, we conclude that footfall does have a strong impact on retail sales and therefore integrating footfall measures into sales forecasting can significantly improve the forecasting results. We also examine differences between the two retailers and observe a stronger correlation at the Fast Food Retailer locations compared to the correlation at Family Restaurant locations.