Are we more forgiving of an artificial intelligence (AI) agent than a human when we are let down? Less appreciative of an AI bot than a human when we are helped? In a new Journal of Marketing study, our research team examined these questions and discovered that consumers respond differently to favorable and unfavorable treatment at the hands of an AI agent versus another human. Our results suggest that AI agents may partially bypass consumer resistance to unfavorable treatment. In addition, AI agents are given less credit for doling out favorable treatment.
Consumers and marketing managers currently are in a period of technological transition where AI agents are increasingly replacing human representatives. AI agents have been adopted across a broad range of consumer domains to handle customer transactions, including traditional retail, travel, ride and residence sharing, and even legal and medical services. Given AI agents’ advanced information processing capabilities and labor cost advantages, the transition away from human representatives for administering product and services is expected to continue. However, what are the implications for customer response and satisfaction?
Our research team found that when a product or service offer is worse than expected, consumers respond better when dealing with an AI agent. In contrast, for an offer that is better than expected, consumers respond more favorably to a human agent. For example, Uber uses an AI machine learning system to estimate travel elasticities and administer ride price offers. Imagine a customer who pays for an Uber ride downtown, but for the return trip unexpectedly receives a price offer triple the original. Our work reveals that administration of this worse-than-expected offer by an AI agent, rather than a human agent, yields better purchase likelihood, customer satisfaction, and intentions toward the use of Uber in the future. However, if the return trip price is unexpectedly much better than anticipated, the response is more favorable when administered by a human.
This effect happens because AI agents, compared to human agents, are perceived to have weaker personal intentions when making decisions. That is, since an AI agent is a non-human machine, consumers typically do not believe that an AI agent’s behavior is driven by underlying selfishness or kindness. As a result, consumers believe that AI agents lack selfish intentions (which would typically be punished) in the case of an unfavorable offer and lack benevolent intentions (which would typically be rewarded) in the case of a favorable offer.
Our team also discovered that designing an AI agent to appear more humanlike can change consumer response. For example, a service robot that appears more humanlike (e.g., with human body structure and facial features) elicits more favorable responses to a better-than-expected offer than a more machinelike AI agent without human features. This occurs because AI agents that are more humanlike are perceived to have stronger intentions when making the offer.
What does this mean for marketing managers? For a marketer who is about to deliver bad news to a customer, an AI representative will improve that customer’s response. This would be the best approach for negative situations such as unexpectedly high price offers, cancellations, delays, negative evaluations, status changes, product defects, rejections, service failures, and stockouts. However, good news is best delivered by a human. Unexpectedly positive outcomes could include expedited deliveries, rebates, upgrades, service bundles, exclusive offers, loyalty rewards and customer promotions.
Managers can apply our findings to prioritize (vs. postpone) human to AI role transitions in situations where negative (vs. positive) interactions are more frequent. Moreover, even when a role transition is not entirely passed to an AI agent, the selective recruitment of an AI agent to disclose certain negative information could still be advantageous. Firms that have already transitioned to consumer-facing AI agents, including the multitude of online and mobile applications that use AI-based algorithms to create and administer offers, also stand to benefit from our findings. Our research reveals that AI agents should be selectively made to appear more or less humanlike depending upon the situation.
For consumers, our findings reveal a “blind spot” when dealing with AI agents, particularly when considering offers that fall short of expectations. Indeed, our findings introduce an ethical dilemma around the use of AI agents – is it appropriate to use AI to bypass consumer resistance to poor offers? We hope that making consumers aware of this phenomenon will improve their decision quality when dealing with AI agents, while also providing marketing managers techniques, such as making AI more humanlike in certain contexts, for managing this dilemma.
Go to the Journal of Marketing