Data mining for automated evaluation of sales opportunities

In this competitive and unforgiving market, reliable financial and business forecasts have become more important than ever. A typical enterprise uses sales opportunity data to make these forecasts. The forecasts are used to predict the firm's future financial performance and to develop plans to achieve them. The accuracy and reliability of these predictions depend on the quality of the data and the techniques used to make them. Inaccurate and unreliable predictions can result in unrecoverable financial losses.;Human experts usually evaluate sales opportunities without significant assistance from decision aid tools. Human errors, shortage of skills, and evaluation costs are thus the major concerns with the evaluation process. These concerns make sales opportunity evaluation an excellent domain for data mining research.;In this thesis, a multi-stage multiclass cost-sensitive classification model is developed using real-world sales opportunity data. The classification model is prototyped and validated with the real-world data and is shown to perform at par with human experts. Special considerations are given to improving its understandability and reducing its sensitivity to minor data changes. Additional experiments show that the multi-stage learning system cannot be replaced with a single stage learning system.;Human evaluations of sales opportunities can be sub-optimal because the factors that influence the final outcome of sales are not visible to the human experts after they successfully exit the evaluation process. Hence, another classification model is trained using real-world data from the sales execution process. The new system classifies enterprise sales opportunities based on their final outcome, i.e. ground truth, rather than on human evaluations. The new classifier provides actionable knowledge for improving human performance within the sales opportunity management process. To better understand their characteristics, winning opportunities were analyzed using clustering. The internal structure of clusters is extracted as a set of rules that can be easily understood and validated by human experts.;The major contributions of this thesis are the formulation of a complex business problem, development of a solution and validation with real-world data. The business implications of this thesis are also discussed. Areas for further research are identified.