In a world going towards automation, I ask whether salespeople making pricing decisions in a high human interaction environment such as business to business (B2B) retail, could be automated, and under what conditions it would be most beneficial. I propose a hybrid approach to automation that combines the expert salesperson and an artificial intelligence model of the salesperson in making pricing decisions in B2B. The hybrid approach preserves individual and organizational knowledge both by learning the expert's decision making behavior and by keeping the expert in the decision making process for decisions that require human judgment. Using sales transactions data from a B2B aluminum retailer, I create an automated version of each salesperson, that learns the salesperson's pricing policy based on her past pricing decisions. In a field experiment, I provide salespeople in the B2B retailer with their own model's price recommendations through their CRM system in real-time, and allow them to adjust their original pricing accordingly. I find that despite the loss of non-codeable information that is available to the salesperson but not to the model, providing the model's price increases profits for treated quotes by as much as 10% relative to a control condition, which translates to approximately $1.3 million in yearly profits. Using a counterfactual analysis, I also find that a hybrid pricing approach, that follows the model's pricing most of time, but defers to the salesperson's pricing when the model is missing important information is more profitable than pure automation or pure reliance on the salesperson's pricing. I find that in most cases the model's scalability and consistency lead to better pricing decisions that translate to higher profits, but when pricing uncommon products or pricing for unfamiliar clients it is best to use human judgment. I investigate different ways, including machine learning methods, to model the salesperson's behavior and to combine salespeople's expertise as reflected by their automated representations, and discuss implications for automation of tasks that involve soft skills.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D83X9Q5M |
| Date | January 2018 |
| Creators | Karlinsky Shichor, Yael |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0073 seconds