Suggested reading:
Why does the New Arrivals algorithm show the old season items that are no longer new?
The New arrivals algorithm sorts the items based on their first visit date with the Click-Stream data collection model. If there is an XML integration, the algorithm refers to the item's addition time to XML. On both models, it is possible to re-upload the old item to the website after Insider's code is integrated, or the old season items could be added to the XML.
To prevent this issue, new information could be added to the XML or collected from the website to ensure that only the new items are presented.
Why do I need to wait to use the Chef Algorithm?
Chef uses big data to train its model to predict the best algorithms in terms of conversion rate performance. The model requires one month of data detailing users' actions across four distinct weekends. It takes at least two months to generate the Auto-Optimized recommendation strategies, allowing for a comparison over the two months and the identification of the trend line in the conversion performance.
When I test the campaign from incognito windows, I always see the same results for the User-based Algorithm. What is the reason for that?
User Based recommendations rely on browsing history, which is tracked using the unique UserID generated by Insider. This UserID is specific to each browser and device. When browsing in incognito mode, a new UserID is generated because no browsing history is stored. Consequently, user-based recommendations resort to backup strategies based on website performance, such as displaying the most popular items, which are the same for all users.
Why do view together and purchase together algorithms show unrelated products?
Those algorithms are not designed to show related items, as their logic is to display items viewed/purchased in the same session, regardless of the relationship. You can add a dynamic filter via the Recommendation Settings if you want to show related items on the product detail pages (like those from the same brand or category).
Can I use more than one algorithm in one variation?
Yes, you can. If you need to use multiple algorithms in the same variation, you can pick the Mixed Strategy and create a bundle of the algorithms.
Can I customize the Smart Recommender algorithms?
Smart Recommender algorithms use automated calculations that cannot be customized. However, you can add different filters to get results based on your preference. For example, you can exclude some categories or boost some brands.
Can I use these Smart Recommender algorithms in different products?
Yes, you can use them for Web Push, Email, InStory, Architect, and Mobile App.
Which algorithms can I use as a publisher?
Four algorithms—most popular items, viewed together, new arrivals, and manual merchandising — are available for Publishers.
How can I check the accuracy of an algorithm?
The algorithm's accuracy is determined by examining the number of user events and the number of eligible products in the catalog. For further information, refer to Monitor Algorithm Health.
Can I shuffle the recommended products for all algorithms?
You can shuffle the recommended products for all algorithms except for the Chef. Since Chef's main feature is determining which algorithm in which order brings the most sales, views, clicks, and other metrics, shuffling the order of the algorithms would render this feature of Chef ineffective.
Why do I receive no feed error despite using manual merchandising?
Ensure that you are writing the correct item_id.
You need to ensure that the products you provide in the manual merchandising are in stock in the product catalog feed. You can verify this on the Catalog Management page.
Is it possible to use the CRM data to train the algorithms?
Yes, it is. You can do it via Upsert API integration. For other technical configurations, reach out to the Insider One team.
What is the difference between trending products and most popular products?
The most popular products are sorted based on their page views for the last 30 days. Trending products are sorted based on the weekly score change trend, which is calculated based on page views and purchase counts.
Why can I not test the attribute affinity?
The attribute affinity needs big data, as does the User-based algorithm. Since there is no way to collect historical data on incognito pages, it is impossible to experience the affinity of the attribute during testing. Additionally, if you are not interacting with the website as a real user (I.e., Purchasing and visiting categories with consistency), the model cannot be trained effectively for you, and therefore, you cannot experience the actual performance of attribute affinity.