Algorithm Reliance Under Pressure: The Effect of High Customer Loads on Service Workers
September 29, Wednesday, 11:30 am - 12:30 pm, on Zoom. The Zoom link is circulated separately via the BEE email list. Please fill the registration form to join our mailing list.
Presenter
Doctoral Candidate at the Stephen M. Ross School of Business
Abstract
Increasingly, companies invest in algorithms that help their employees serve customers effectively and efficiently. Algorithms can rapidly analyze information about a customer’s past preferences and turn this into a recommendation for how to serve this customer at the present. Because algorithms can analyze customer information faster (and good algorithms, more accurately) than humans alone, one advantage for companies is that servers can manage higher customer loads (but only if servers actually follow algorithms’ recommendations). In this paper, we study the effect of high and low customer loads on servers’ algorithm reliance with a laboratory experiment. Our experiment design is based on an algorithm implemented at the academic coaching company ReUp, our research partner. We find that while people are generally reluctant to rely on the algorithm’s recommendations about how to serve a customer, reliance is significantly higher for high versus low loads (47.7% versus 38.5%, respectively). This difference exists partly because high load servers rely more on the algorithm over time – while low load servers do not – because high load servers see more feedback about customer satisfaction for recommendations made with and without the algorithm. Consequently, servers learn the algorithm is better than expected. Although a higher customer load is more difficult to serve, high load servers achieve levels of customer satisfaction on par with low load servers because of their greater algorithm reliance. Our findings have implications for companies about how to design and implement human-plus-algorithm service models.