top of page

Generative AI Lifts Online Retail Sales by Up to 16% in Field Tests

3 hours ago
2 min read

Image via Wikimedia Commons (public domain).

Does generative AI actually raise sales, or only productivity on paper? A 2026 Management Science study ran large-scale randomized field experiments on seven GenAI workflows at a leading cross-border online retailer. Most workflows increased sales, by up to 16.3%, mainly through higher conversion, with no rise in returns or drop in ratings.

Seven workflows, one retailer, randomized evidence

Much of the evidence on generative AI at work comes from lab tasks, surveys or single deployments. This study takes a different route. According to Fang, Yuan, Zhang, Donati and Sarvary (Management Science, 2026), a leading cross-border online retail platform integrated GenAI into seven business workflows during 2023 and 2024, and the authors evaluated these rollouts through a series of randomized field experiments involving millions of users and products.

The workflows span the customer journey and the seller side: customer service, matching consumers to products, advertising, and services for sellers. Because each application was tested with randomization, the study can speak to causal short-term effects on sales rather than correlations between AI adoption and performance.

The headline is heterogeneity. GenAI adoption increased sales in most workflows, but effects ranged from no detectable impact to 16.3%. The authors link this spread to GenAI's marginal contribution relative to what the firm was already doing: where existing practices were strong, there was less room for the AI to add value. Across the four applications with positive sales effects, they estimate an annual incremental value of roughly $5 per consumer, which they describe as economically meaningful given the retailer's scale and the early stage of adoption.

The mechanism points to frictions. Gains came mainly from higher conversion rates rather than larger cart values, which the authors read as GenAI reducing search, information, communication and personalization frictions in the shopping experience. Two further results matter for anyone worried about AI pushing low-quality sales: product return rates and customer ratings did not deteriorate. The benefits were also larger for less experienced consumers.

The study measures short-term effects at a single platform during an early phase of GenAI adoption, so the size of the effects may change as tools mature, as competitors adopt similar features, and as consumers adjust. Readers should consult the paper and its online appendix for workflow-level details, since an average across seven different applications hides very different designs.

Why it matters: For marketing and information systems researchers, this is rare large-scale causal evidence on GenAI's commercial value, and it shows why evaluating 'GenAI' as a single treatment is misleading: the effect depends on the workflow and on the baseline it replaces. It also offers a template for measuring AI impact on outcomes such as conversion, returns and ratings, and a clear prompt to study who benefits, since less experienced consumers gained most.

Fang, L., Yuan, Z., Zhang, K., Donati, D., & Sarvary, M. (2026). Generative Artificial Intelligence and Sales Productivity: Field Experiments in Online Retail. *Management Science*. Read the original. Independent editorial summary of third-party research.

Comments


This initiative is supported by the following organizations:

  • Twitter
  • LinkedIn
  • YouTube
logo_edited.png
bottom of page