| Abstract: | To meet sales targets with finite resources, business-to-business (B2B) sellers need to prioritize promising sales opportunities from a large pipeline of possibilities. Extant evidence in theory and practice suggests that B2B sellers use arbitrary or gut-based decision rules to prioritize sales opportunities, which leads to sub-optimal sales opportunity management. I draw from the relationship management and organizational buying literature to build a conceptual framework relating buyer-class typology (e.g., new bid vs. modified rebid), and opportunity characteristics (e.g., opportunity size) to a seller⁰́₉s decision to bid, and the bid outcome. I test the framework using archival data from a major B2B on-site services provider. The data span sales outcomes (did not bid, bid and won, bid and lost) for 4,564 sales opportunities, spanning 10 years, and 23 countries. Model-free evidence indicates an ongoing tension in sales pursuit: while the best projects to bid on are ones with the best relationship typology (e.g., direct relationship) and lowest risk (e.g., small opportunity size), such opportunities are not enough to grow the sales in each region. Accordingly, I first develop an ensemble machine learning framework to predict the focal seller⁰́₉s propensity to win the opportunity, as a function of complex non-linear interplay among the sales opportunity characteristics. Subsequently, I propose to build a combinatorial optimization approach that allows a company to take the right level of risk (i.e., some new bids, some large opportunities) so that each region can maximize its sales potential given finite capacity. I demonstrate that using the ensemble approach enable the focal company to improve its predictive validity over simple models by 11.09%. Moreover, by using a combinatorial optimization approach in conjunction with the predictive ensemble model, I retrospectively demonstrate that the focal firm could have increased its sales by 22% while bidding on 38% fewer projects. The electronic version of this dissertation is accessible from https://hdl.handle.net/1969.1/197943 |
|---|