When will workers follow algorithm advice? Retailers introduce advanced algorithms to optimize decisions such as pricing, inventory, and product assortments. However, it is not effective if workers do not follow the algorithm. How can we motivate workers following the algorithm? Who is more or less reluctant to follow the algorithmic advice?

In the beverage vending machine business in Japan, we conducted a field experiment to answer these questions. We provided algorithm advice on product assortment decisions at the vending machines and tested whether workers change their behavior. The workers were incentivized to increase sales. Nevertheless, they were not willing to follow the advice. They suggested that the increased burden of work was the main reason for the nonconformity. High-performing workers were more likely to ignore the advice. However, if we ask for workers’ predictions on the sales of new products and integrate them into the initial value of the algorithm, some workers followed the algorithm, and, in this case, high-performing workers were more likely to follow.

The result suggests that the nominal performance of an algorithm is not fully achieved in reality unless the manager successfully convinces workers to follow the algorithmic advice. For doing this, incorporating the worker’s opinions and involving them in the design of the algorithm can be one of the solutions, especially for motivating high-performing workers.