Retailers already collect and analyze sales transaction data to better understand their stock position, inventory turns, store promotions, and customer promotions. In the future retailers are also likely to collect data about product returns, customer inquiries, price checks, customer complaints, customers that visit a store but do not buy anything, and a customer’s browsing behavior in the store. In fact, electronic storefronts can already capture much of this data. Customer interaction data is sales data plus all other data related to a customer’s total relationships with a retailer. Customer Interaction Mining is the analysis of customer interaction data to garner insight about a retailer’s customers.

Customer Interaction Mining research at the HITC attempts to answer the following questions:

• What is the business value of this additional data?
• What research problems does customer interaction data create?

Two research problems are created by customer interaction data:

1. Structured Data--Customer interaction data are not all collected at the same level of detail. For example, sales data is usually collected by product by store. Customer complaints data, on the other hand, might relate to a group of products or no product at all. In order to include customer complaints data in an analysis, the data analysis algorithm should be able to understand the structures between the different types of data and take advantage of these structures.
   
2. Temporal Data--When analyzing sales transaction data, the sequence of purchases is usually ignored. However, with customer interaction data the sequence of sales, browsing, complaints and returns cannot be ignored, because this sequence contains insights into customer behavior.

Technologies being investigated at the HITC for Customer Interaction Mining are Bayesian Belief Networks, Case-Based Reasoning, Hidden Markov Models and the Minimum Description Length Principle