Incorporating e-commerce into retail location models

The use of location models in retail businesses is well-established, particularly in the grocery sector. Many alternative methods are in use today but the spatial interaction model (SIM) has a proven record of success. To date, that success relates purely to face-to-face activities, modeling and predicting visits by consumers to retail outlets. However, grocery retailers are cutting back on store investments and concentrating on investment in the convenience market and e-commerce: the latter has now reached a 7.2% share of the U.K. grocery market, with continued growth forecast. Although spatial models are used extensively for helping to locate new convenience stores, so far e-commerce has not been built into existing retail location models. Yet e-commerce seems to be a spatial activity. Extensive evidence demonstrates the geography of demand and supply are as important in groceries e-commerce as they are in face-to-face grocery retailing. We therefore take up the challenge of incorporating e-commerce into classic location models. Methodologically, we find the standard distance deterrent term in the production-constrained SIM unsuitable for modeling e-commerce flows: we explore inverting this term and find extensive gains in prediction accuracy, an interesting finding that contributes to the ongoing applied SIM literature.