Breed better banners: Design automation through on-line interaction



The aim of this article is to explore the use of an automated design system based on a Genetic Algorithm to design banner advertising. Just as genes encode traits in living organisms and propagate the most successful traits by natural selection, so this algorithm propagates features of the most successful banner advertisements and extinguishes the least successful traits. In the current experiment at a major on-line store selling CDs, the algorithm interactively evolved 320 different Web-banners over a period of 23 days. Although the algorithm exhibited some deficiencies, the results indicate that the method is viable. It created innovative designs that performed increasingly better than reference banners, improving the click-through rate from 0.68% for a standard banner to 3.1% by the 16th generation of design, without human intervention from the advertiser. The article concludes with a discussion of some conceptual issues and directions for future research. © 2002 John Wiley & Sons, Inc. and Direct Marketing Educational Foundation, Inc.