The transformation isn’t flashy. There’s no humanoid robot behind the screen, no futuristic shopping assistant making suggestions or buying for you. But in one of the world’s largest online marketplaces, Generative AI has been quietly reshaping how millions of consumers find, evaluate, and buy products, producing significant economic gains.
Columbia Business School Professor Dante Donati, alongside coauthors Lu Fang, Zhe Yuan, Kaifu Zhang, and Miklos Sarvary, partnered with a leading global e-commerce platform to uncover what GenAI is really worth to business performance. Their findings go beyond hype or speculation to offer real-world evidence that AI can make firms more productive by improving how people shop.
AI That Sells
Between September 2023 and June 2024, the researchers tested seven AI-driven workflows — from smarter search and chatbots to automated product descriptions and marketing messages — across millions of users and products. Each was run as a randomized field experiment, randomly assigning consumers or products to AI-enhanced or standard experiences. Prices, staffing, and other inputs stayed fixed, ensuring any increase in sales reflected real productivity gains.
The results were striking. Across the workflows, GenAI lifted sales by up to 16.3 percent, with the strongest effects in customer service and search.
The AI-powered pre-sale chatbot delivered the largest gains: sales rose by 16.3 percent, while conversion rates—how often browsing turned into buying—increased by 21.7 percent. When AI was used to refine search queries, translating and clarifying multilingual input, sales increased by nearly 3 percent as shoppers more easily found what they were looking for. Similarly, when AI generated richer, localized product descriptions, sales rose by about 2 percent, particularly for listings that had previously been sparse or poorly translated.
Aggregated across applications, these effects translate to roughly $4.6 to $5 in added annual value per consumer, a meaningful gain given the platform’s global reach and the early stage of GenAI adoption.
Productivity Without Job Cuts
The researchers found that these gains came without reducing labor or capital. Notably, AI improved productivity by enhancing customer experience, not cutting costs. Chatbots answered questions instantly, refined queries helped users find better product matches, and automated descriptions made listings richer. Conversion rates rose as high as 22 percent, indicating smoother shopping journeys and more completed purchases, with no corresponding rise in average cart value
AI didn’t make work faster, but instead made the marketplace smoother. That means the technology makes true total factor productivity gains, or more output with the same inputs.
Still, not every application delivered results. When AI was used to generate Google search ad titles, researchers found a statistically insignificant decline in sales, likely because the model wasn’t fine-tuned for e-commerce and produced suboptimal titles.
Who Wins Most
AI’s benefits weren’t evenly distributed, but they were equitable. Smaller and newer sellers, typically constrained by limited marketing resources or staff, saw larger gains than established vendors.
On the consumer side, newer or less active shoppers benefited the most. In the chatbot experiment, conversions among inexperienced users jumped 26%, compared to about 20% for veteran buyers. In the search refinement and product description experiments, sales among inexperienced buyers increased by up to 8%, while they did not change for more tenured shoppers.
By improving matching, and closing gaps in information, experience, and access, AI helped both sides of the market catch up, a contrast to past technological shifts that often widened inequalities. The researchers noted that AI appears to act as a market equalizer, narrowing outcome gaps between small and large players.
