Amazon introduces Amelia, an AI assistant for third-party sellers

Amazon is rolling out an artificial intelligence tool designed to help third-party sellers quickly resolve issues with their accounts and fetch sales and inventory data.

The company said Thursday that it’s launching the product, called Amelia, in beta for select U.S. sellers, before introducing it more broadly later this year. Amazon describes it as an “all-in-one, generative-AI based selling expert,” and is making it accessible through Seller Central, the internal dashboard for third-party merchants.

Amelia is the latest generative AI tool that Amazon has brought to market in the past year as it seeks to capitalize on the hype sparked by OpenAI’s ChatGPT. The company has introduced an AI-powered shopping assistant named Rufus, a chatbot for businesses dubbed Q and Bedrock, a generative AI service for cloud customers.

Amazon also plans to upgrade its Alexa voice assistant with generative AI features, CNBC previously reported, and the company has invested billions of dollars in OpenAI competitor Anthropic, its largest venture deal to date.

CEO Andy Jassy told investors earlier this year that the “generative AI opportunity” is almost unprecedented and that increased capital spending is necessary to take advantage of it.

“I don’t know if any of us has seen a possibility like this in technology in a really long time, for sure since the cloud, perhaps since the internet,” Jassy said on the company’s first-quarter earnings call in April.

Google and Microsoft have introduced rival products to try to ensure their relevance in a market that’s predicted to top $1 trillion in revenue within a decade.

AI has also become more prevalent across Amazon’s e-commerce platform. The company now displays AI-generated summaries of product reviews and it’s launched AI features for third-party sellers that can help them write listings and generate photos for ads.

Amazon also said Thursday it’s launching tools that let sellers create AI-generated video ads and use AI to write product listings in bulk based on their entire catalog. The company said it’s beginning to use generative AI to show personalized product recommendations and listings based on a user’s shopping history. For instance, Amazon would show the term “gluten free” in the description for a box of cereal if a shopper typically searches for products with that phrase.

Amazon made the announcements at its annual conference for sellers hosted in Seattle. Third-party sellers are the heartbeat of Amazon’s dominant e-commerce business. Since about 2017, they’ve accounted for at least half of all goods sold on the site. In the second quarter of this year, that number swelled to 61%.

Dharmesh Mehta, Amazon’s vice president of worldwide selling partner services, told CNBC in an interview that a growing number of merchants are using its AI services. More than 400,000 of Amazon’s millions of third-party sellers have used its AI listing tool, up from 200,000 in June, he said.

With Amelia, Amazon is counting on generative AI to help with a key issue for third-party merchants — account troubleshooting. The company has sprawling teams that help sellers resolve account suspensions and deal with inventory issues, as well as build their business on the site. Merchants have long complained about the difficultly with getting swift resolution or reaching a human when unforeseen issues surface with their accounts.

The company said Amelia can offer help investigating an account issue and, in the future, will be able to “solve the problem on the seller’s behalf.” Mehta described how instead of filling out a form for missing inventory, a seller could ask Amelia to file a claim for them or the tool could resolve the issue automatically.

“There are going to be places where, hey, instead of chatting with seller support or getting on the phone with someone, maybe Amelia is able to do that and do that faster,” Mehta said. “I don’t need to send an email to someone and wait for a response.”

Amazon said Amelia uses Bedrock, a software tool that lets users access large language models from Amazon and other companies like Anthropic and Stability AI. Mehta said Amelia is trained on public data from the web, along with information pulled from Amazon seller resources, FAQs and other public-facing websites.

Mehta said the model isn’t trained on seller-specific data, which is closely guarded.

Amazon said the tool uses retrieval-augmented generation, or RAG, a popular AI industry framework that combines generative AI with long-established methods of information retrieval. It allows the pulling of certain seller-specific information from Amazon’s internal systems without storing it or including it in model training data.