Abstract:
Based on a survey among 22 direct marketing experts and in-depth project experiences within automotive industry, the authors identify two key issues: first, when the success of direct marketing campaigns should be measured; and second, when the results of a campaign should be used for selecting the target group for a new campaign. As a solution to the latter, i.e. too few cases for learning a stable and accurate scoring model, we propose the semi-supervised modeling approach when selecting learning cases: an initial classifier is learned on the ideal participants from a previous campaign. This classifier is employed to select supplementary cases. This is a research technique that has been applied in other domains for several years, but thus far not in (direct) marketing.