Job Market Paper
"Zero to One: Sales Prospecting with Augmented Recommendation" (with Saiquan Hu, Juanjuan Zhang)
Helping new salespeople succeed is critical in sales force management. We develop a recommender system to help new salespeople identify customers with better conversion potential. One challenge is how to deal with salesperson-customer combinations that have no historical sales records. These instances are treated as missing observations in standard recommender systems. We instead consider the possibility that sales records are absent because previous salespeople were unwilling or unable to sell to certain types of customers. We develop a parsimonious model to capture these endogenously absent sales records and embed the model into a neural network structure to form an augmented recommender system. We validate our method using sales force transaction data from a large insurance company. Our method outperforms popular industry benchmarks in prediction accuracy and recommendation quality.