Automatic customer targeting: a data mining solution to the problem of asymmetric profitability distribution

This paper proposes a data mining approach for automatic customer targeting based on their expected profitability. The main challenge with customer profitability prediction is asymmetry, i.e., skewness of the distribution, because the number of highly profitable customers is very small compared to others. Although data mining methods are more resistant to sample heterogeneity than statistical ones, due to strong skewness, the accuracy of predictions often decreases as the value of profit increases. These few customers are actually outliers which can make data-driven methods to overestimate predicted amounts, but on the other hand, they contain very important information about the most valuable customers, so it is not advisable to remove them. In this paper, a data mining approach for overcoming these problems is proposed. The results show that the relative error in predicting the absolute amount of the profitability of the most valuable customers is very small and does not differ much from the error for other customers, unlike previously applied methods where predicting high profitability was less accurate. Accordingly, the specific implication of the high accuracy is more efficient identification of the most profitable customers, which ultimately make a greater contribution to the company in terms of revenue. Also, due to the good precision of the model, errors in the assessment of highly profitable and risky customers are reduced, which leads to savings in unnecessary costs for the marketers.