A program called "Innovative Billing" has been developed to provide individualized energy information for a mass audience - the entire residential customer base of an electric or gas utility. Customers receive a graph on the bill that compares that customer's consumption with other similar customers for the same month. The program aims to stimulate customers to make ef?ciency improvements. To group as many as several million customers into small "comparison groups", an automated method must be developed drawing solely from the data available to the utility. This paper develops and applies methods to compare the quality of resulting comparison groups. A data base of 114,000 customers from a utility billing system was used to evaluate Innovative Billing comparison groups, comparing four alternative criteria: house characteristics (floor area, housing type, and heating fuel); street; meter read route; billing cycle. Also, customers were interviewed to see what forms of comparison graphs made most sense and led to fewest errors of interpretation. We find that good quality comparison groups result from using street name, meter book, or multiple house characteristics. Other criteria we tested, such as entire cycle, entire meter book, or single house characteristics such as ?oor area, resulted in poor quality comparison groups. This analysis provides a basis for choosing comparison groups based on extensive user testing and statistical analysis. The result is a practical set of guidelines that can be used to implement realistic, inexpensive innovative billing for the entire customer base of an electric or gas utility.