A Multiagent System Approach to Grocery Shopping

Abstract

We present an approach to social grocery shopping based on customers trading information about item prices and quantities in order for the customers to find the lowest prices for the goods they purchase and the most convenient plan for buying them. Because collecting and reporting prices is tedious, agents repre-senting customers are needed to make this approach practical. Agents also have the potential to learn which other agents can be trusted. We use a realistic shopping list based on the U.S. Consumer Price Index in order to guarantee the realism of our results. By visiting actual grocery markets and comparing prices, we have discovered that the total cost of a list of groceries can vary by 13%. We have also discovered that by shopping optimally, that is, buying each item from the cheapest store, the result can be a savings of 16% over shopping at the store with the lowest total cost. To shop at minimum cost requires customers’ agents to report prices to each other. If they do, each customer is likely to achieve at least a 10% savings. However, what if the reported prices are inaccurate? Would customers be worse off than if they just shopped randomly? We investigated the robustness of our multiagent shopping system in the presence of errors in reported prices. From this, we determine the potential savings an average customer might obtain.