Optimising retail networks: A case study of locating Camelot's lottery terminals

Abstract

Camelot has run the National Lottery since its inception in November, 1994. In summer 1998, Camelot decided to investigate whether its existing network of retail outlets within which its 24,500 online terminals were located, was in fact the optimum network — or whether it would be possible to make improvements. This paper sets down the background to Camelot's retail operation, and how the investigation was conducted. It then goes on to show how a spatial interaction (gravity) modelling approach was devised and implemented, with the aim of providing Camelot with a mechanism to optimise future terminal placement. Although gravity modelling is not new this particular optimisation model was unusual in several respects. First, against the scale of task. In the more conventional modelling that might be used in (say) grocery retail, a spatial interaction model would typically be used to support decisions relating to a single, or small numbers of possible, predetermined locations. The Camelot model uses a genetic algorithm that allows optimisation across entire regions of the UK at a time, and involves the model working across many thousands of possible retail outlets and millions of possible resulting networks. Secondly, against the usual demand flows (from residential areas to a retail destination) the Camelot model allows for more complex flows (from home, work or shopping trip). Finally, the paper explains how the model was operationalised within a geographic information system, and implemented for Camelot to run in-house.