A well-structured and data-enriched feed sits at the core of a successful recommendations program. How you build and manage your feed is the most significant factor in creating effective recommendations, and provides the opportunity to leverage business use cases that are otherwise not available. Based on years of experience with customers, we've identified some best practices for several industries and verticals, and include them here for your benefit.
Categories column (mandatory)
This column includes all the categories associated with an item, separated by pipes. This is one of the most important columns of the feed, and it directly affects the performance of many algorithms and the ability to merchandise and filter recommendations at scale.
For this reason, it's extremely important to ensure the quality and accuracy of the categories hierarchy. The key to creating the best category list is to eliminate the marketing categories from the natural category tree, and include only those categories relevant to placing the item in the right place in the store.
Good hierarchy: Women|Trousers & Jeans|Chinos
Bad hierarchy: Women|Summer Collection|Trousers & Jeans|Buy 1 get 1|Chinos
Pro tip: For advanced use cases, we highly recommend also adding each category from the Categories column as a separate column of its own, so that your marketers can use it in various merchandising scenarios and advanced recommendation widgets.
So, if we take the given example, you could create the hierarchy Women|Trousers & Jeans|Chinos in your Categories column, and then create separate columns for 1st-level, 2nd-level, and 3rd-level items, which would respectively include Women, Trousers & Jeans, and Chinos as values. This would enable you to create a strategy that matches Women, but not Trousers & Jeans.
Key words are an array of data attributes, separated by pipes. They describe the product and its features. Although not mandatory, a rich keywords array significantly improves the system's ability to generate strong Similarity models and strong and accurate Affinity profiles, and we highly recommend creating one.
We highly recommend that each keyword also be added as a separated column so that your marketers can use the keywords in merchandising efforts.
Examples for e-commerce:
- Material (polyester, cotton)
- Color (red, blue, black)
- Product Style (skinny, printed, denim, oxford)
- Product Collection (Disney, Star Wars, cartoons)
- Body Fit (Tall, Plus Size)
- Age Group (Baby, Kids, Adults, Elders)
Examples for consumer packaged goods (CPG):
- Package size
- Dietary Restrictions
- Sale Item
Include this information when weather conditions or the season can be attributed to an item, and this data is available. Examples can include soup or umbrellas for winter, camping or picnic items for summer, rakes for autumn, and so on. An example of how this information can be used might include an international brand that sells items in multiple countries, with an offering that includes products for multiple weather conditions or seasons. Having weather granularity included in the product feed enables the brand to adjust recommendations results to match user needs based on regional forecasts.
Sales and sale badges
Sale information is used to enable advanced merchandising rules, such as the prioritization or deprioritization of sale items. In many cases, the "sale" badge is an integral part of the recommendation widget's look and feel. If your Dynamic Yield campaign is designed to serve both the widget and its container, this information must be included in the product feed.
Product categories and product brand URL
In some advanced recommendation use cases, you might want to refer the customer to the brand page or one of the category pages associated with a recommended product. For example, showing the customer a notification with a 'shop more from this brand' button for a recently purchased item - clicking the button will lead to the purchased item brand page.
Pair It With, Shop the Look, and other dynamic bundles
Your product feed can also accommodate unique business logic developed by your merchandisers or master stylists, the use of custom columns. Learn more
Diet, health, and lifestyle (CPG)
While for most e-commerce businesses the user’s affinity profile is focused on interactions with products and their categories, this is not the case for most CPG businesses. In a grocery scenario, you probably won't find shoppers with an exceptional affinity to the Dairy category. But you probably do want to aim your recommendations to users' dietary restrictions, like gluten-free, vegan, or kosher/halal. This granularity in your product feed enables you to seamlessly favor products that are aligned with users' special requirements.
Available online only (CPG)
Hybrid businesses that operate brick-and-mortar stores together with online shopping might not have the same item availability in both channels. To enable recommendations of relevant in-store and online-only items, and in some cases also display the information as a badge, you must include this information in your product feed.
Package units (CPG)
Most CPG businesses, especially groceries, recognize different persona types, identified by their affinity to the package size and units they consume. For example, B2B vs. B2C, families vs. singles, and so on. Having the unit granularity in the feed (for example, 6-pack vs. 4-pack) enhances affinity collection. We recommended creating a column that clusters the various packages into human-readable groupings such as “Small” “Medium”, and “Large” packages.
Private label (CPG)
When constructing your feed, flag your private label products (or other sponsored labels) that you might want to exclude from certain slots or pin into certain slots, regardless of other factors, like product popularity.