From Aversion to Adoption: The Role of Promotion Design in Mitigating Algorithm Aversion

The views expressed in this article are the sole responsibility of the authors and cannot be attributed to Compass Lexecon or any other parties.

Abstract

Despite rapid advances in artificial intelligence (AI), adoption is often hindered by algorithm aversion, a psychological reluctance to trust algorithmic systems. Overcoming this barrier is critical for realizing the potential of AI innovations. While prior research has primarily focused on system-based solutions, we examine a widely adopted economic incentive, price promotions, whose effectiveness in reducing algorithm aversion has yet to be examined. In collaboration with a leading global e-commerce platform, we conducted a field experiment involving 26,276 existing platform users who had never previously used autonomous delivery vehicles. Participants are randomly assigned to a control group, a single-stage promotion (one-time offer) group, or a multi-stage promotion (a series of discounts) group. Our results show that while promotions trigger initial trials for autonomous delivery, they also risk cannibalizing full-price orders during the promotional period. Nevertheless, customers who received coupons exhibit a significant increase in full-price adoption of autonomous delivery relative to the control group after the promotions end. Furthermore, we find the single-stage strategy spurs a trial but cannibalizes full-price sales and has no lasting impact. By contrast, the multi-stage strategy drives adoption more than five times greater than the single-stage, generates positive spillovers to full-price orders even during the promotional period, and fosters sustained long-term use. Its effectiveness follows an inverted U-shaped, peaking at an average coupon interval of 11 days and a discount depth of 51%. Subsequent heterogeneous analyses uncover the patterns behind these divergent outcomes. Customer habituation to self-pickup tends to undermine the limited benefits of the single-stage strategy, while it amplifies the effectiveness of the multi-stage strategy. Moreover, the effectiveness of both strategies is also contingent on service reliability, as delivery failures ultimately undermine their efficacy. Overall, our study provides a scalable blueprint for designing promotions to build trust, reduce algorithm aversion, and sustain the adoption of AI technologies.