1. Go to Content > Recommendations from the left navigation. The Recommendations page is displayed.
2. Click Create recommendations. The Create recommendation page is displayed.
   
   
3. Select the relevant model according to your use case and click Next.
   
4. In the Recommendation name field, enter a name for the recommendation.
5. In the Recommendation description field, enter a description for the recommendation
6. In the Catalog drop-down list, select the relevant catalog from which you wish to retrieve these recommendations.
   Note: All AI recommendations consider one catalog. After you use a catalog in any AI recommendations, it will be automatically configured for all other types of AI recommendations.
7. Choose the relevant user actions based on your use case from the drop-down list.
   • In the case of Similar items, Frequently bought together items, and Frequently viewed together items, the recent most item based on the configured user action events will be used to generate.
   • In the case of Recommended items, you are not required to configure any user actions. All performed user actions by a user are considered to learn their behavior patterns automatically and the recommendations are generated tailored to the shifting interests of individuals.
8. Click Save. A recommendation is created with a 'Processing' state on the Recommendations page. It can take a maximum of 24 hours for recommendations to be active.
   
9. After the recommendation is ready, you can test results for sample users before using it in any campaign.

Best Practices

• Choose the global catalog inclusive of all the business offerings or the type of offerings you want to generate recommendations for.
• Map all relevant events in the relevant user action buckets.
• Ensure that the unique item ID attribute has the same information as the unique item ID in the catalog.
• Configure all possible user-item interactions in the Recommended Items settings, such as Product Viewed, Added to Wishlist, Added to Cart, and Product Purchased. This allows the algorithm to learn from a wide range of user behaviors and adapt to rapid changes.
• Test the results for sample users before incorporating them into your campaign.