COSMO: how Amazon’s new artificial intelligence works

Understanding user intentions by analysing big data from e-commerce platforms can be very useful in various applications, such as correctly guiding product searches. From a cognitive point of view, there is really nothing new, or almost nothing: while it is true that what we most easily see are behaviours, it is also true that it could be much more useful and beneficial to go back and intercept intentions, which are the mental states on which those same behaviours depend.

Now then, it is not necessary to be great experts to understand how important the analysis of intentions can be, as it can fuel more personalised and effective search and usage experiences on platforms.

Well, it is in this scenario where artificial intelligence can add value. But how?

A first experiment was carried out using the largest language models such as GPT and OPT, asking them why users buy certain products. The result of this utilisation is the possibility of obtaining common-sense knowledge with a fair degree of accuracy and low costs.

However, as can easily be imagined, this experiment is far from infallible. So much so that, according to the tests carried out, the models do not manage to give a precise explanation. In some cases, they simply respond that consumers have bought certain products because they like them. Or, perhaps, that an Apple Watch was bought simply because it is a watch.

The arrival of COSMO: what changes?

Based on these premises, Amazon has been working hard to find useful solutions that revolutionise and rethink the way we find products on its platform. The latest system, in terms of time, has been named COSMO and, exactly as the first researchers into the application of artificial intelligence to e-commerce had supposed, it seeks to understand the intention behind the search and what the customer really needs.

But what concretely changes with COSMO?

The American company itself suggests this, as on the 10th of May it released more information about its framework, which uses large language models to generate a coherent knowledge graph from customer interaction data. This graph can include and encode the relationships between products and the human contexts in which they are used, such as their functions, the recipients, and the locations.

COSMO foresees, in other words, a recursive approach in which a Large Language Model (LLM) generates hypotheses about the common-sense implications of request-purchase and co-purchase data. The hypotheses are then filtered using human annotations and machine learning models.

In summary, COSMO is the tool with which Amazon aims to make searches smarter, sifting through the product catalogue and finding items that best match the search intentions. The goal is evidently to avoid searches providing results that are inconsistent with the potential buyers’ purposes. This is something that, however, can often happen with the current search approach. In a shared example, for COSMO, the search for slippers for pregnant women involved finding non-slip slippers.

What changes for sellers with the introduction of COSMO?

Evidently, the arrival of COSMO brings quite a few new developments for sellers. Sellers are asked to take some precautions that will allow their products to improve visibility in the eyes of those searching for solutions to their needs.

In reality, these are not true revolutions in the approaches that should already have been of interest to sellers. However, the launch of COSMO will be an opportunity to focus even more attention on these aspects. Let’s try to summarise them below.

Measure the real profit from your Amazon sale with the ZonWizard Profit tool

Be clear and transparent with buyers

Try to use product descriptions that speak directly and simply to consumers, highlighting the most peculiar and real characteristics. All the information you consider important should be communicated openly. This will allow Amazon to qualify the product more accurately. As a result, the chances of it appearing in more reliable search results will increase.

In summary, it is always worth focusing deeply on product information, ensuring that your listings include detailed information about:

  • Product functions
  • Intended recipients
  • Contexts of use

This will help COSMO better understand your product and recommend it to the right customers.

Personalise the listings

Amazon provides its sellers with a lot of information about customers’ search and purchase habits. Additionally, you can obtain an even greater level of informative depth by using the suite of services found in ZonWizard. Use the data to customise the listings according to what people seem to want.

Adapt the listings

According to the information released, COSMO is expected to act continuously to refine its approach. For this reason, sellers must be ready to quickly adapt to its evolutions.

Encourage customers to leave detailed reviews

It is very likely that reviews will become increasingly important for COSMO’s learning mechanism. For this reason, and to develop a more favourable visibility of the product, it is useful to encourage buyers to leave detailed reviews.

Incidentally, you can also do this with our suite. Have you already started the free trial?

Research keywords

This advice is not exactly new. However, nowadays it requires a slight adjustment due to Amazon’s new artificial intelligence model. In other words, it has become essential to conduct a search to find the keywords that best align with the common-sense relationships identified by COSMO. Keywords should be included in the product listings. This way, the chances of products appearing in relevant search results can be increased.

Categorise products

In addition to keywords, categories are also an element that cannot be underestimated. In other words, the precise categorisation of products is fundamental for COSMO in order to generate meaningful relationships. It is important to ensure that products are in the correct categories within Amazon’s product graph.

Think about customers’ needs

Finally, the last piece of advice, and perhaps the most important. When creating your product listings, you should always put yourself in the shoes of your current and potential customers. Ask yourself what problems they are trying to solve, and how and where they might use your product.

Try the Profit Tool for free

Take full control of your Amazon Business. You will see with clarity real profits, margins, ROI, orders, costs and have a balance for each European marketplace. If you are not yet familiar with ZonWizard's Profit Tool, find out how it can help you manage and grow your business.

Learn more

2 thoughts on “COSMO: how Amazon’s new artificial intelligence works

Leave a Reply