Precision

The Precision setting allows you to choose the types of recommendations that are allowed to be shown in a given profile. In some cases, you may want strict control over which recommendation types are shown, opting to show fewer results or even hide the profile if your preferred results are not available. In other cases you may want to fall back on some broad, generic result types to make sure something is shown in that location, even if the preferred type is not available.

The options for this setting range from Broad to Narrow.

• A broader precision will allow recommendations to “fall back” in situations where data is insufficient, so that some alternative form of recommendations can be presented to the shopper. 
• A narrower precision will restrict the profile so that recommendations only show if they meet the criteria for the selected type. The narrower setting may result in the profile not showing on certain pages when the minimum threshold of products has not been met.

Profiles default to the broadest precision, but can be changed and made more restrictive when needed.

Note: This feature is exclusive to Similar, Cross-Sell, and Cart-Cross-Sell recommendations.

Our data science generator analyzes user behavior for each item in the catalog. It also analyzes how those items are related by catalog attributes such as categories, collections, and brands, and how shoppers interact with products owning those attributes.

From this analysis, we are able to generate crowd data for each item (bought together or viewed together), and also genomic data (bought with items like this or viewed with items like this). 

Supported Precisions

• Curated will limit recommendations to the products set in Product Rules. If the seed product does not have one or more products associated with it in a rule, no recommendations will be returned.
  • If merchandisers want to create a "Complete the Look" experience, they would set the Precision Filter to Curated. As a result, the recommendations carousel would only show the pinned products.
• No Fill will only apply the crowd and genomic results for the requested type.
• Crowd backfills with alternative recommendations based on all shopper activity (similar backfills to Cross-Sell and vice-versa). No genomic recommendations are shown.
  • Once a recommendation is selected, we will show crowd data for that recommendation, and backfill with crowd data of the alternative. 
  • The Data Science Generator will analyze user behavior to determine popular cross sell patterns. As an example, people who bought "this shirt" often bought "those pants". With Crowd Precision, the term "those pants" becomes a popular cross sell recommendation for "this shirt".
• Genomic backfills by incorporating recommendations for other products that share product attributes with the one being viewed (helpful when a product is new). 
  • We will show crowd and genomic for the selected recommendation and can backfill with the crowd or genomic of the alternative.
  • Let's return to the example of "this shirt" and "those pants". If the Data Science Generator can not find enough data regarding what people bought with "this shirt", it will generate recommendations based on the shirt's attributes. So, people who bought "other products similar to this shirt" also bought "that jacket". 
• All will incorporate Crowd and Genomic fallback recommendations, and will also add top-selling products (category and site-wide) if no better recommendations can be found.
  • This is the default configuration and it includes the entire hierarchy mentioned above. 
  • Merchandisers can still curate recommendations and they will show up first. However, with "All" Precision, the AI will fill in the rest at all the different precision levels. 

The table below lists all the recommendations that are made when you use Curated, No Fill, Crowd, Genomic, or All precision.