Skip navigation

RUFUS, COSMO, Amazon and AI: how online retail is changing in 2024

– Written by Jess Chapman

Robot with laptop

Image by catalyststuff on Freepik

Amazon employs some of the world’s most famous uses of artificial intelligence (AI). 

From Alexa’s conversational capabilities, to the handy AI-generated review summaries on product listings, Amazon has long been an AI pioneer. 

And it’s not showing any sign of slowing down, with a couple of major developments in 2024 that will revolutionise how online retail works - by reading customer intent, and matching it to suitable products. 

Let’s explore what they are and how they work.

Introducing RUFUS, the AI shopping assistant

First of all, there’s RUFUS. RUFUS is an AI shopping assistant for customers, launched in beta this year.

RUFUS is trained on Amazon’s product catalogue, customer reviews, community Q&As, and information from across the web to give customers conversational answers to their questions.

Customers can ask questions like “What should I consider when buying headphones?” or “What’s the difference between bone conduction headphones and in-ear monitors?” or “Can you recommend waterproof IEMs?” 

RUFUS will generate useful answers, which shoppers can respond to with follow-up questions. It should make online shopping easier for customers.

We predict as RUFUS grows, it will become more aware of the context and customer intent of Amazon’s products, so will be able to give more accurate answers and recommendations.

And that’s where COSMO becomes important. 

What is COSMO?

COSMO is Amazon’s new Large Language Model (LLM) AI system that reads customer intent, beyond keyword matching.

What does that mean? It means COSMO will find contextual, common-sense detail in descriptions and in the images themselves, and match products to the customer’s intent. 

The creators of COSMO say it will encode “relationships between products in the Amazon Store and the human contexts in which they play a role – their functions, their audiences, the locations in which they’re used.”

COSMO is very new: it was presented as a paper at a data conference only a couple of months ago. But it’s likely to change everything we currently understand about product search, as it becomes semantic-based, rather than keyword-based. 

Give us an example? 

Yes, the best way to explain is through examples. So for example, you might currently be selling a jacket on Amazon using the keyword “windbreaker jacket”. 

But thanks to COSMO, Amazon’s product search is becoming more based on semantics - that is, language and meaning - and will be able to understand how your jacket can be used. So if a customer searches “jacket to climb Mount Kilimanjaro”, it’s more likely to match with your product, despite you not using that keyword.

Image by Amazon

Here’s another example. A search for “shoes for pregnant women” can be enhanced by COSMO’s knowledge that pregnant women require slip-resistant shoes. It is more likely to turn up a slip-resistant shoes ASIN in the results, even if this ASIN has no ‘shoes for pregnant women’ keywords. 

It means that brands, who are used to thinking about keywords that best describe the product, should start to start to think more about their audience’s intent in using the product. 

Tell me more - how can I enhance my content for COSMO? 

You’ll need to ensure your titles, descriptions and bullets include some intent data to match customers’ wants and needs. 

Look at the example below - the words in green are intent content. COSMO is most likely to match these words with what it knows about what customers are searching for. 

It should be pretty obvious that those intent words by themselves would bring up a whole range of irrelevant products. But the intent data can make the search much more specific and the results much more relevant. 

Source: Keplo

You mentioned COSMO searches images - how does that work? 

COSMO is smart enough to read your images and pick up intent clues from them. 

For example, if you use lots of photos of your product being used by a certain demographic - let’s say pregnant women again - COSMO will assume the product is intended for that audience. 

It can also spot other objects in your images, such as a scoop or glass of finished smoothie in the background, and use that information too. 

“We’re slowly but surely transitioning from keyword search to semantic search. Where the search was influenced by mostly product attributes, it is now moving more into intent and use cases around products. Exciting times.”

- Claudiu Clement, CTO, e-Comas

What can I do to get ready for COSMO?

You’re probably already using plenty of intent words in your titles, bullets, descriptions and A+ content. 

But we highly recommend shaping your content around your products’ intended uses going forwards, because it looks very likely that semantic-based search will be here to stay. 

This doesn’t mean keywords will be redundant. Take another look at the example customer intent phrase above (“for digestion, immunity, energy”) – it’s pretty useless information without the main keywords. So for now, at least, don’t let your keyword research slip.

We recommend A/B testing your titles at the very least over the next few months to see if more intent words yield better results, particularly if you’re selling in the US, where RUFUS and COSMO are being rolled out first. Keplo.com is great for intent research. 

We also recommend bearing COSMO in mind when you’re planning new product imagery. Don’t forget Amazon has a new genAI image enhancement tool that can create ‘lifestyle’ pictures of your product in contextual use. 

Context will be everything in product search, so make sure you take advantage of this if you’re short on graphics and resources. 


Want to talk to a human about Amazon’s AI solutions? Give us a call today!