Amazon embraces Vendors to improve AI merchandising
It’s a funny world we live in. Amazon sends an email to a pilot group of vendors and says, “Hey, we’re excited to let you know we improved your product listings with our AI engine. The changes are not published yet, but you have until February 7th to either accept or reject the changes and if you do nothing, we will automatically publish the improvements on February 7th.”
Nice. I’m guessing this is part of the continuing development and expanded use of Amazon Bedrock and who knows how many of their 2 million Vendors received the invite. Unfortunately, GenAI has the ability to be grossly inaccurate and it’s up to the Vendors to catch the errors. Here are a few real world examples:
· AI: change the product material from copper to steel. Copper is a soft non-sparking, non-magnetic metal. Steel is not. It can throw off friction sparks, very magnetic and hence very dangerous in high risk combustible or magnetic environments.
· AI: change the product description from uncoated to powder coated. Some metal products are powder coated for protection of the bare metal. Other metal products, like those made of copper or bronze are inherently corrosion resistant and do not need a protective coating. In fact, the users want to see and work with bare metal – uncoated metal.
· AI: change the size of a socket from 9/16 to 1-9/16. Adding an inch to the size of a wrench that is 9/16, really doesn’t help the customer when they need a 9/16” wrench. Just because AI thinks it should be an inch bigger doesn’t mean it should be an inch bigger.
Those are three simple examples, but just think of what this will do for anything where size is important: clothing, tires, tools, raw materials, electronics, etc.
And what about the legal risks when AI comes along and changes a safety product into something that is very unsafe. Does Amazon get in trouble and stuck with the legal risk? Probably not since it is commonly the responsibility of the manufacturer to ensure Amazon has correct and accurate product information.
The reality is, Amazon just enlisted up to 2 million business owners that will manually review, check and double check their AI recommendations on millions of products. And what does it cost them? Nothing. Zippo. Nothing out of pocket, but certainly an improvement to their AI merchandising model.
And what happens if a few thousand business owners don’t read the email and review the AI recommendations? Well, let’s just say, buyer beware, because you just don’t know what the AI agent did to the product description and whether it is right or wrong.
A few months back I made note of a very useful GenAI Governance Framework. In this Framework, the authors talk about Transparency, Accountability and Continuous Improvement. In particular the following recommendation is made:
“Ensure transparent and traceable AI decision-making”
As, I think about Amazon and their use of AI to improve merchandising within their massive multi-sourced product catalog, I have to wonder how or IF they will highlight for shoppers the millions and millions of changes made by AI in describing products posted on their web site. I also, have to wonder how they can so casually change the product description of so many products when Vendors (Amazon owned inventory) and 3rd party Sellers (Seller owned inventory) alike are dependent upon the same product descriptions. Sure it’s only a pilot today, but wow, what an audacious move!