India is increasingly becoming a testing ground for Amazon’s next generation of AI tools. The country’s scale, diversity of users, and mix of first-time and sophisticated online shoppers create conditions where product challenges surface early and solutions, once tested, often travel globally.
Rajeev Rastogi, vice-president–machine learning (ML) at Amazon, who oversees ML technology for emerging markets, told FE that many innovations originate in India because problems tend to be sharper and more visible here.
2% Background Lift
One example is Amazon’s use of generative AI to enhance product images. Indian customers in categories such as apparel found plain white backgrounds less appealing. Amazon experimented with AI-generated neutral-tone backgrounds that preserved realism, including shadows and depth.
“We did an experiment and we saw that the sales went up 2% from just changing the background colour for these categories like apparel, certain categories and so on,” Rastogi said.
India has also driven advances in catalogue automation. Because many sellers in emerging markets are less digitally equipped, Amazon developed AI systems that extract product attributes — such as colour, sleeve type or collar style — directly from images and limited text inputs. The approach, first prioritised in India to address catalogue-quality gaps, is now influencing tools across other marketplaces.
Customer-facing AI experiences have followed a similar path. Amazon built troubleshooting chatbots in India to help customers resolve post-purchase issues, reflecting the country’s preference for assisted shopping similar to the offline store guidance. These capabilities are now integrated into Rufus, Amazon’s AI shopping assistant, and used in other regions as well.
Beyond Search
The company has also used India’s heterogeneous user base to refine adaptive interfaces and personalised search. By analysing behaviour during a session, Amazon can classify shopper proficiency and adjust the interface accordingly.
“The more proficient customers, see ads, different widgets, subscribe and save options, Prime (offerings), and so on. For the less proficient users, there are tutorials, different language options and simplified navigation for instance,” Rastogi said.
Search results are likewise tailored using signals such as regional product popularity, brand affinity and price sensitivity, techniques first tested where such diversity is most pronounced.
GenAI is also reshaping product videos. Amazon already uses AI to stitch together images, feature highlights, scripts and voiceovers into short clips that explain product benefits. While fully automated, studio-quality video from simple prompts is still evolving, the company expects rapid improvements as models mature.
Underlying these efforts is access to usable data — and the constraints around it. Rastogi noted that ML teams build on whatever datasets are cleared for use, while separate compliance teams determine what can be accessed under country-specific regulations. That means experimentation can continue, but launches may be delayed or redesigned depending on evolving rules.