Recently, there has been a lively discussion about Amazon's new algorithm, COSMO. This morning, a classmate told me that there have been significant changes in BSR recently. Is it related to the algorithm changes?
Amazon has indeed been testing a new algorithm recently. They have used approximately 10% of the traffic in the United States for A/B testing and achieved a 0.7% increase in sales. At the same time, the participation rate in the search navigation bar has increased by 8%.
Coincidentally, I have also roughly read the content before, so today I will briefly talk about my thoughts.
Download COSMO Algorithm Source File
Table of Contents#
01 / Conclusion
02 / Interpretation of COSMO Algorithm
03 / Seller Optimization Direction
01 / Conclusion#
- New traffic entrance
- Some impact on popular old links
- Attribute segmentation, vertical products have new opportunities
The A9 algorithm is still an important foundation for Amazon, and AI algorithms are only a part of it. Overall, COSMO is beneficial for new products.
After reading the whole article, it seems a bit like Taobao's personalized recommendations, but it is a bit "rudimentary" because Amazon allows customers to make their own choices through COSMO's multi-round navigation and Rufus's conversational search recommendations, and only aims to make the search results more accurate. It cannot achieve the level of personalized recommendations based on customer tags like Taobao.
However, this may increase the time cost for customers, so it is currently uncertain whether buyers will be accustomed to or accept this shopping habit. If customers can accept the new algorithm, the traffic of some "popular" products will decrease, and more interesting products with specific features will be discovered to obtain more orders.
02 / Interpretation of COSMO Algorithm#
Mining User Intent#
![Mining User Intent.jpg]
For example, if a pregnant woman wants to buy a pair of shoes and searches for "pregnant shoes" in the Amazon search box, the COSMO algorithm will analyze the importance of features like "anti-slip" for pregnant women based on a large amount of "human common sense" stored in language models such as GPT or OPT, as well as by asking about "reasons for user purchases or common purchases." Then, it recommends products that can capture the pain points of pregnant women's needs, such as anti-slip shoes, to improve conversion rates.
Multi-round Navigation#
![Multi-round Navigation.jpg]
For example, a search for "camping" may lead to the selection of "inflatable mattresses" and then further refine to "inflatable mattresses for camping." COSMO will then provide various types of inflatable mattresses for camping to meet different needs, such as lakeside camping, mountain camping, or camping for 4 people.
Amazon Search Knowledge Graph#
![Amazon Search Knowledge Graph.jpg]
After looking at the above image, we can see that these contents are very similar to the audience targeting in SD advertising. They start from different directions such as functionality, scenes, audience, seasons, and interests.
In the future, we can consider adding or giving more consideration to these factors when writing listings.
Session-based Shopping#
This feature should refer to Rufus, which was exposed some time ago. It helps customers search, discover, research, and purchase products through "question and answer" interactions.
Rufus has entrances in the Amazon app's top search box and product detail pages. For example, when I enter the detail page of a dress and I don't feel like reading the details, I can directly ask if this dress is suitable for someone who is 1.8 meters tall. Rufus will provide answers and recommendations based on buyer reviews, Q&A, listing information, and more.
03 / Seller Optimization Direction#
- Carefully explore the usage scenarios and target audience of products, and provide detailed descriptions of product features and usage scenarios in listings.
- Explore more segmented vertical markets and provide products that meet customers' personalized needs, which can effectively avoid homogeneous competition.
- Use AI tools for listing optimization, making it easier for COSMO to find and recommend our products based on AI logic.
- Establish a good independent website and "disguise" our brand as a trustworthy one, making it easier for COSMO to discover our products when retrieving data.
Amazon often iterates its algorithms, but some updates may not be released. However, regardless of the changes, the customer-centric philosophy will never change.