Modelling Spatiotemporal Fluctuations of Consumer Demand in the UK Grocery Sector and their Impact on Retailers Store Sales

Retail location planning in the grocery sector is used to aid location-based decision-making; addressing issues such as, store performance, store planning or for estimating demand. Location planning teams employ a range of sophisticated modelling techniques; this thesis draws on Spatial Interaction Models (SIMs). Spatiotemporal components have received limited attention in relation to retail location planning and this is addressed within this thesis. It is argued that spatiotemporal components of consumer demand fluctuate considerably within catchment areas, which has a direct impact of store sales, and that models, which have accounted for specific aspects of spatiotemporal demand result in better and more representative store revenue estimations. This thesis demonstrates the importance of spatiotemporal demand through an analysis of store level diurnal sales patterns which are related to observed consumer shopping habits, providing novel insight and improving our understanding of supply and demand-side spatiotemporal components. The insight is used to generate a series of spatiotemporal demand layers, which are used in conjunction with a (SIM) to generate spatiotemporally informed store level revenue estimations. Rare access to temporal EPOS transaction records at a store level and substantial loyalty card data provided by a major player in the UK grocery market presents novel opportunities for analysis. The data enables this thesis to generate insight into temporal fluctuations of store sales and the demand side drivers behind them, as well as, the geographies of consumer demand. The findings demonstrate evidence of distinct clusters of diurnal sales profiles in stores, which appear to be directly related to specific spatiotemporal demand types, and is used in conjunction with a SIM. The analysis shows that through adding spatiotemporal demand layers, demand estimates within catchment areas are far more representative and can lead to improved revenue estimations.