Enhancing the Prediction of Missing Targeted Items from the Transactions of Frequent, Known Users

The ability for individual grocery retailers to have a single view of its customers across all of their grocery purchases remains elusive, and is considered the “holy grail” of grocery retailing. This has become increasingly important in recent years, especially in the UK, where competition has intensified, shopping habits and demographics have changed, and price sensitivity has increased. Whilst numerous studies have been conducted on understanding independent items that are frequently bought together, there has been little research conducted on using this knowledge of frequent itemsets to support decision making for targeted promotions. Indeed, having an effective targeted promotions approach may be seen as an outcome of the “holy grail”, as it will allow retailers to promote the right item, to the right customer, using the right incentives to drive up revenue, profitability, and customer share, whilst minimising costs. Given this, the key and original contribution of this study is the development of the market target (mt) model, the clustering approach, and the computer-based algorithm to enhance targeted promotions. Tests conducted on large scale consumer panel data, with over 32000 customers and 51 million individual scanned items per year, show that the mt model and the clustering approach successfully identifies both the best items, and customers to target. Further, the algorithm segregates customers into differing categories of loyalty, in this case it is four, to enable retailers to offer customised incentives schemes to each group, thereby enhancing customer engagement, whilst preventing unnecessary revenue erosion.

The proposed model is compared with both a recently published approach, and the cross-sectional shopping patterns of the customers on the consumer scanner panel. Tests show that the proposed approach outperforms the other approach in that it significantly reduces the probability of having “false negatives” and “false positives” in the target customer set. Tests also show that the customer segmentation approach is effective, in that customers who are classed as highly loyal to a grocery retailer, are indeed loyal, whilst those that are classified as “switchers” do indeed have low levels of loyalty to the selected grocery retailer. Applying the mt model to other fields has not only been novel but yielded success. School attendance is improved with the aid of the mt model being applied to attendance data. In this regard, an action research study, involving the proposed mt model and approach, conducted at a local UK primary school, has resulted in the school now meeting the required attendance targets set by the government, and it has halved its persistent absenteeism for the first time in four years. In medicine, the mt model is seen as a useful tool that could rapidly uncover associations that may lead to new research hypotheses, whilst in crime prevention, the mt value may be used as an effective, tangible, efficiency metric that will lead to enhanced crime prevention outcomes, and support stronger community engagement.

Future work includes the development of a software program for improving school attendance that will be offered to all schools, while further progress will be made on demonstrating the effectiveness of the mt value as a tangible crime prevention metric.