Predicting Resource Usage for Capital Efficient Marketing

A structural conflict exists in businesses that sell services whose production costs are discontinuous and whose consumption is continuous but variable. A classic example is in businesses where capital-intensive infrastructure is necessary for provisioning service, but the capacity resulting from capital outlay is not always fully and efficiently utilized. Marketing departments focus on initiatives that increase infrastructure usage to improve both customer retention and ongoing revenue. Engineering and operations departments focus on the cost of service provision to improve the capital efficiency of revenue dollars received. Consequently, a marketing initiative to increase infrastructure usage may be resisted by engineering, if its introduction would require great capital expense to accommodate that increased usage. This conflict is exacerbated when a usage-enhancing initiative tends to increase usage variability so that capital expenditures are triggered with only small increases in total usage. A data warehouse whose contents encompass both these organizational functions has the potential to mediate this conflict, and data mining can be the tool for this mediation. Marketing databases typically have customer data on rate plans, usage, and past responses to marketing promotions. Engineering databases generally record infrastructure locations, usages, and capacities. Other information often is available from both general domains to allow for the aggregation, or clustering, of customer types, rate plans, and marketing promotions, so that marketing proposals and their consequences can be evaluated systematically to aid in decision making. These databases generally contain such voluminous or complicated data that classical data analysis tools are inadequate. In this article, we look at a case study where data mining is applied to predicting capital-intensive resource or infrastructure usage, with the goal of guiding marketing decisions to enable capital-efficient marketing. Although the data mining models developed in this article do not provide conclusive positions on specific marketing initiatives, and their engineering consequences, the usage revenues, and infrastructure performance predicted by these models provide systematic, sound, and quantitative input for making balanced and cost-effective business decisions.