Behavioral Externalities of Process Automation

We study the behavioral effects of process automation on human workers interacting with automated tasks. We introduce a stylized normative model with two workers who complete their tasks sequentially, working toward a joint project to obtain a fixed payment plus a variable bonus that depends on how early the project is completed. The normative model prescribes that, if workers are fully rational, they will complete their tasks as soon as possible if the early completion bonus is high enough. However, following the literature, we hypothesize that workers will suboptimally delay project completion. Following the insights from a behavioral model, we further predict that automation will alleviate this problem by reducing strategic uncertainty, resulting in an indirect behavioral benefit of a higher worker productivity, in addition to the direct benefit of a higher project completion rate and a shorter project duration. To test these predictions, we conduct an experiment replicating the theoretical model, varying (i) whether a worker collaborates with a coworker or a robot, and (ii) in the case of collaborating with a robot, whether the upstream or downstream task is the one automated. First, we find that workers largely deviate from the optimal policy, as they take longer than what the normative theory prescribes to complete their tasks or do not complete the project. Second, we show that process automation increases the project completion rate and reduces the project completion time, confirming the benefits of process automation. Interestingly, workers who collaborate with robots take longer to complete their tasks, contradicting our initial hypothesis that process automation has a positive effect on the productivity of human workers. In addition, we find that upstream automation is more beneficial than downstream automation. We also show that social preferences are an important driver of these results because prosocial subjects tend to be more productive when collaborating with a human coworker than with a robot. Finally, we show that our findings remain robust in a continuous processing setting.