Events

How should firms deploy AI alongside human workers as AI systems advance from assistive tools to autonomous agents? This talk presents evidence from a research collaboration with Alibaba on AI deployment in customer service operations, anchored by two complementary field studies.

The first study, “Generative AI in Action: Field Experimental Evidence from Alibaba’s Customer Service Operations,” examines how a gen AI assistant—providing issue diagnosis and solution suggestions to human chat agents—affected worker performance in after-sales service. Gen AI improved service speed and subjective quality measures such as customer ratings, but the gains were strikingly heterogeneous: low performers improved the most, narrowing the performance gap, while top performers experienced declines in both subjective and objective quality. The evidence points to an unintended behavioral mechanism: top performers leaned more heavily into multitasking and shifted away from concurrent chats more often, slowing their responses to customers. The second study, “Agentic AI and Human-in-the-Loop Interventions: Field Experimental Evidence from Alibaba’s Customer Service Operations,” moves a rung up the autonomy ladder—from AI that assists human agents to agentic AI that resolves customer chats on its own. It examines when human-in-the-loop interventions rescue AI failures and when they do not, and finds that human oversight preserves quality in technical escalations but is far less effective in emotional escalations.

Taken together, the two studies suggest that the operational and behavioral consequences of AI deployment depend critically on the level of AI autonomy and on which workers, customers, and task types are involved. I will close by drawing out implications for how firms should design hybrid human–AI service systems and identifying open questions for the next wave of research on AI in the workplace.