Ai Assistance, Employee Creativity, and Job Performance: Evidence from a Field Experiment

Abstract: Can artificial intelligence (AI) create complementarity with human employees by assisting them at work, and if so, how and when? We examine AI assistance in a sequential division of labor within organizations, wherein AI handles the initial portion of a task that is repetitive and well-codified so that employees can focus on subsequent higher-level problem solving. This organizational design can help overcome both AI’s limitations in performing higher-level problem solving and humans’ boredom with repetitive work. We provide causal evidence from a field experiment in a telemarketing firm which randomly assigned AI assistance to generate sales leads in the first stage (a well-codified and repetitive portion of the sales task) and then handed over the leads to human employees for sales persuasion in the second stage (which required higher-level problem solving). We find that AI assistance can significantly increase employee creativity in answering customers’ questions, but only for top employees serving difficult customers. Moreover, AI assistance increases employee efficiency only for bottom employees serving easy customers. In both cases, AI assistance increased employee performance. We thus highlight employee capabilities and customer types as critical scope conditions for attaining complementarity between AI and humans in problem solving. Policy simulation results suggest that the AI-human team can be optimized by matching AI assistance with the right employees and the right customers. These findings offer useful implications for organizations to generate augmented intelligence and achieve optimal performance from AI-human collaborations.