Identifying & Targeting Prospective Buyers

Goal: Merge Information onto a National Household Database for a Direct Mail Campaign

By: Bob Ceurvorst, Decision Systems Group, Market Facts, Inc.

Market Facts Inc. (http://www.bloomberg.com/profiles/companies/1655066D:US-market-facts-inc) is an international custom market research consulting company specializing in collecting and processing information to assist clients with strategic and tactical business decisions. The company designs, executes, and interprets market research conducted on behalf of its clients for applications ranging from new product ideas to brand loyalty. Their clients include most of the largest 100 multinational consumer products and services companies, as well as business-to-business firms and other organizations.

Market Facts conducted a large, multi-country project for a client interested to determine who are the best prospects for purchasing an in-vehicle navigation system (IVNS) with their next new car or truck, and why. Realistically, most people are unlikely to buy this type of product in the near future, so segmentation is necessary to identify likely prospects. Since few vehicle manufacturers have offered IVNS’s in the past (although the number is growing) and few customers have purchased them, current owners are not representative of potential future owners. For these reasons, prospects were identified via a segmentation (cluster analysis) of consumers based on their responses to 30 attitude statements covering their driving patterns, competence and comfort using maps and directions, and opinions (pro and con) about IVNS’s. Five segments were identified, of which two were good target markets: “Road Warriors” who drive a great deal, often in unfamiliar areas, and “Security Seekers” who need directional help and view IVNS’s as providing welcome safety. The same five segments emerged in an independent analysis of European consumers giving additional credence to the cluster structure.

In order to target likely Road Warrior and Security Seeker households throughout the U.S. via direct mail, it was necessary to develop an algorithm to identify them using information available in a national 100-million+ household database. Using only data available in the national household database, WizWhy yielded rules that classified 80% of the surveyed U.S. households correctly as in or out of the Road Warrior group (recovering all of the Road Warriors) and 84% correctly as in or out of the Security Seeker group (recovering most of the Security Seekers).

To compare WizWhy’s performance versus the more commonly used decision tree approach, an experienced data miner at the company housing the national household database used decision trees to develop an algorithm to predict Road Warriors. After several runs, the best result he obtained was two splits, resulting in too few ending nodes and naturally, poor predictive accuracy.

WizWhy is especially useful in cases where the number of records available for analysis is not large. Decision trees are hampered by the fact that (a) the small initial sample is carved into even smaller pieces as the analysis progresses and (b) the resulting rules are all nested. The discovered rules in a decision tree with k levels would certainly appear in the WizWhy rule set allowing up to k predictors in a rule. However WizWhy produces many more useful and statistically significant rules, all derived using all of the records.