Project Nile: Amazon's Secret AI Plan to Change How You Shop … – Business Insider

Jump to

Earlier this year, Amazon VP Joseph Sirosh showed up for an internal meeting wearing an “I love AI” t-shirt with a heart symbol.
A former AI executive from Microsoft, Sirosh had joined Amazon in late 2022. Now he had an important message for colleagues: The Amazon e-commerce experience is getting a massive makeover.
Insider obtained a recording and transcript of his presentation. Other internal documents and interviews with people familiar with the matter reveal a radical overhaul by Amazon that will change how consumers shop online.
“We have a big task ahead of us,” Sirosh said. “We need to orient everyone towards this new world.”
The giant retailer is revamping the search experience on its website and app to weave in generative AI capabilities that are more conversational, in-depth, and personalized.
Codenamed Project Nile, the initiative aims to add a layer of artificial intelligence on top of the existing search bar on Amazon, enabling instant product comparisons, requests for more details and reviews, as well as recommendations based on search context and personal shopping data.
The new search was supposed to launch in September, but got delayed, one of the people said. It’s being tested internally, and could launch in January, starting with the US, another person said, though those plans could change again. The sources asked not to be identified because they’re not authorized to speak to the press.
“Project Nile is a confidential initiative wherein we’re building a conversational shopping agent for Retail customers,” one internal document explained.
AI-powered, interactive search could improve the shopping experience on Amazon. The company is also hopeful that the changes increase sales, especially on mobile devices. The project is a top priority internally with backing from senior leaders, including CEO Andy Jassy and retail boss Doug Herrington, the people said.
The project has broad implications because Amazon is the default gateway for many online shoppers. More than 60% of US consumers start product searches there, according to e-commerce software maker Jungle Scout.
“While we don’t comment on future plans, we’re going to keep inventing in generative AI across all of our businesses,” an Amazon spokesperson wrote in a statement.
Amazon employees got a glimpse of Project Nile at an internal town hall meeting, titled, “Reinventing Amazon Shopping with LLMs,” held earlier this year.
Wearing that “I love AI” t-shirt, Sirosh, VP of Amazon Search and Alexa Shopping, introduced Project Nile as a “super confidential” project. He described it as “AI transforming shopping” and “bringing in the power of large language models into understanding customer questions and answering them.”
The new AI assistance layer will proactively provide search results, expert answers, and product suggestions.
For example, if a shopper asks Amazon’s new search bar, “What kind of coffee maker should I get?” it will give multiple options for a drip, pod, or espresso coffee machine, eliminating the need to visit each product page one at a time. If the shopper follows up by asking for the best pod machines, it will show the top choices based on different metrics, like the number of positive reviews. The shopper can then compare other features, like brew times or cup sizes, or ask for more details about specific products, such as Keurig pods. It can also provide summarized reviews and personalized recommendations, based on individual order history and wish lists.
Sirosh compared the new AI feature to a deeply knowledgeable in-store salesperson who’s familiar with each shopper’s individual taste.
“Before e-commerce, the salesperson in the store was your search engine. And that individual knew everything about the products…they would look at you and know what you might want because customers like you have been to that store before. They may have known you in person, so may know your preferences, and then they can synthesize all of that information together in natural language conversation with you and help you in your shopping mission,” Sirosh said. “And if you can scale that up to everything in Amazon, that would be the future mission we want to aspire to.”
Sirosh said Amazon plans to add a conversational layer to the currently existing search bar, which means the new AI feature can either expand or go away depending on the user’s search query. The new search feature may not use just one large language model, but multiple models, he added.
It will likely be rolled out to Amazon’s mobile app first because mobile search has a lower conversion to purchases than the desktop website, even though it constitutes nearly 80% of searches on Amazon.
“If we can increase the conversion on mobile, because we provide great experiences, great expert answers, then that could be a potentially significant lift to Amazon,” Sirosh said.
Project Nile isn’t the only move Amazon has made in the popular generative AI space. Last month, the company broadly updated its Alexa voice-assistant with a smarter and more conversational AI model. AWS made its new AI-building platform Bedrock generally available, while committing to invest up to $4 billion in Anthropic, an OpenAI competitor. In July, Jassy promoted SVP and head scientist Rohit Prasad to lead a newly created artificial general intelligence team to build Amazon’s “most ambitious” large language models, as Insider first reported.
“Amazon has been using machine learning and AI for more than 25 years in virtually everything we do. We’re investing in generative AI across all of our businesses,” the company spokesperson wrote in their statement. “In our retail business, over the past few months we’ve introduced customer review highlights generated with AI, as well as new tools that make it easier for sellers to write engaging, effective product listings.”
For Amazon’s retail side, Project Nile is one of the most important projects. Sirosh has told his team he’s “staffing up very rapidly” as Project Nile is the “absolute top priority.”
The concept for Project Nile was one of the many ideas Amazon employees put together earlier this year in an internal document titled, “Generative AI-ChatGPT Impact and Opportunity Analysis,” as Insider previously reported. Amazon has been scrambling internally to find ways to take advantage of the generative AI boom since OpenAI wowed the world with ChatGPT late last year. In August, Jassy said “every single one of our businesses” was working on multiple generative AI projects.
Amazon believes Project Nile has a competitive edge because it will have access to proprietary shopping data from Amazon’s wide-ranging product catalog, along with user behavior data and purchase and review information, according to a person involved in the project.
To further improve the quality of search results, Amazon wants to use human AI trainers who will review the AI-powered answers, this person said. These people will evaluate answer quality and compare that against ChatGPT, Microsoft’s Bing Chat, and Google’s Bard. It could also partner with Alexa for a richer experience that expands the text queries to voice search, this person said.
One of the biggest concerns internally is the accuracy of search results, this person said. AI services from OpenAI, Microsoft, and Google sometimes show erroneous information, commonly called “hallucinations.” Amazon hopes to mitigate those problems by using content moderation tools and human reviewers, or by simply refusing to answer sensitive questions, like those related to healthcare products, the person told Insider.
Amazon may also look to third-party providers for help. It is considering using ChatGPT to provide answers for the more sensitive questions, while it may potentially partner with other AI companies, like YouChat, for more data support, this person added.
During the internal town hall meeting, Sirosh was critical of Amazon’s shortcomings as a search engine. He said Amazon in its current form isn’t able to use past search data to preserve the context in future results because “every search is new” on Amazon.
Given the intensifying competition, Amazon will need to move fast, Sirosh added, though security and compliance issues could make that difficult.
“We are unfortunately going to be restricted from doing things like GPT and OpenAI for data security reasons. Many of our competitors won’t be and they have less to lose,” Sirosh said, according to the transcript obtained by Insider.
Still, Sirosh was confident about AI’s role in dramatically changing Amazon’s search business, and the overall shopping experience.
“Generative AI is the new normal,” Sirosh said at the meeting. “In the world of software, everything that we do is going to fundamentally change.”
Do you work at Amazon? Got a tip? 
Contact the reporter Eugene Kim via the encrypted-messaging apps Signal or Telegram (+1-650-942-3061) or email ( Reach out using a nonwork device. Check out Insider’s source guide for other tips on sharing information securely.
Read next