Configure AI Recommendations Rules

Commerce lets you create complex recommendation strategies in the administration interface. You can manage these rules across widgets on different page layouts.

This section applies to both OSF and Storefront Classic. This section applies to Open Storefront Framework (OSF) and Storefront Classic.

Product recommendations are generated by an AI-powered learning engine. Strategies and global exclusions work in harmony with this engine to provide granular control over the merchandising experience.

  • Recommendation strategies are rules you build to customize the kinds of products that are recommended to shoppers.
  • Global exclusions are rules you build to prevent certain products from ever being recommended to shoppers.

Understand recommendations strategies

Commerce product recommendations are generated by an AI-powered learning engine. The engine analyzes each shopper's behavior in real time to recommend the most relevant products at every interaction.

Commerce provides a number of pre-defined strategies that let you restrict displayed recommendations, for example, to products the shopper has most recently viewed, or products that are most frequently bought by other shoppers who have also viewed a product that a shopper is currently viewing.

In addition to these pre-built strategies, Commerce lets you create your own strategies that provide customized guidance to the recommendations engine. You can use strategies to restrict the kinds of products that are recommended, creating a more curated experience for shoppers and providing the flexibility to tailor your recommendations plan to match your merchandising goals.

Each custom strategy is made up of one or more recommendation groups. A recommendation group is a set of conditions that are evaluated together to specify a group of product recommendations. If a strategy includes multiple recommendation groups, the Product Recommendations widget will display products returned by each recommendation group sequentially, that is, all products returned by the first recommendation group, followed by all products returned by the second recommendation group, and so on.

If a product is returned by multiple recommendation groups in a strategy, it appears only once in the Product Recommendations widget. The product’s display position will be determined by the first recommendation group that returns it.

The following table describes each type of condition you can add to a recommendation group.

Condition type Description
Brand The value of the product’s brand property in the context of the current product or products being viewed. If there are multiple products, such as the products in a cart or a wish list, products in any of the brands can be part of the condition.
Browsed Together Returns recommended products based on the most frequently browsed products by other shoppers associated with the product the shopper is currently viewing.
Collection

The condition returns products based on either the same collection or collections in the context of the current product or products being viewed, or only in collections you specify. If multiple collections are represented, products in any of the collections can be recommended. If there are no collections in context, no restrictions are placed on the recommendations.

A product can be in one or more collections. For more information, see Organize products in collections.

Manually Related Products Returns recommended products based on the products specified as related products for the current product or products being viewed. See Display related products for more information about adding related products to a specific product.
Most Recently Viewed Returns products the shopper has viewed. Recommended products are displayed starting from the most recently viewed. First-time shoppers on the site are offered no product recommendations.
Price Lets you select a price range of products to recommend.
Price Type Lets you select products from specific price list groups to recommend.
Purchased Together Uses the context of the product the shopper is currently viewing. Recommended products are displayed based on the most frequently purchased products by other shoppers associated with the product the shopper is currently viewing.
Top Sellers

Restricts recommendations to global top sellers. If the recommendation group contains other conditions, Top Sellers will be applied after all other rules.

The algorithm for the Top Sellers strategy is based on products purchased. It generally considers the top ten percent of recently-purchased products from the product catalog to offer as recommendations. The strategy does not offer products in a sequential order of most purchased items and has no time restrictions as to when items were purchased. Top Sellers is designed to personalize items recommended to each unique shopper. While a particular product might sell more frequently than other products, that product may not be recommended based on a shopper’s navigation of the site.

Understand product context

Product context lets you decide whether a condition for a recommendation group or exclusion rule should be dynamically inferred from the current page. The following procedure explains the steps Commerce goes through to provide context for recommendations conditions when the shopper is viewing pages on your store:

  1. If the current page has a brand or collection, it is used for the context of Brand or Collection.
  2. Otherwise, if the current page has a product, its brand, parent collections, co-browses, and co-buys are used as the context for Brand, Collections, Browsed Together, and Purchased Together conditions.
  3. Otherwise, if there are current wish lists, the products in the wish lists are used to define the context as described in the previous step.
  4. Otherwise, if an order has been placed in the current session, the products in the order are used to define the context as described in step 2. An order is defined as an item that has been purchased during the shopper’s current session. The item could have been placed in the cart during a past session.
  5. Otherwise, if the shopper has added products to their cart but not purchased them (either in this session or their most recent previous session), the products in the cart are used to define the context as described in step 2.
  6. Otherwise, no context is found, so the current condition will not apply.

Understand recommendations for account-based shoppers

Commerce lets you create accounts for companies that do business with you, such as manufacturers, distributors, and wholesalers. Each account represents a single organization and a unique customer. Shoppers who are associated with an account are called contacts.

The recommendations engine considers the behavior of all contacts who are members of an account when determining which recommendations to show an individual contact. (This also applies to recommendations strategies and exclusions you create.) While recommendations continue to be generated for an individual contact, Commerce also factors in key information about other contacts within the same account. Shared account behavior ensures that buyers now see relevant recommendations based on interaction from all buyers across an account.

For example, suppose buyers within the same account are viewing and purchasing the same products. To enhance the quality of recommendations, the Commerce aggregates information about related purchases and customer behavior to optimize the quality and breadth of products that are recommended.

For more information about working with account-based stores and shoppers, see Configure Business Accounts.

Display recommendations

Once you have created and published custom recommendations strategies, you can customize the Product Recommendations Carousel widget to use specific strategies for different locations on your site. This helps you create more specialized product selections appropriate to each shopper's current context. When you customize the widgets settings on the Design page in the administration interface, you can select any custom strategies you published, along with the out-of-the-box strategies. For more information, see Display product recommendations.

Create recommendation strategies

To create a recommendation strategy, follow these steps:

  1. On the Marketing page, click Strategies under AI Recommendations.
  2. Click the New Strategy button.
  3. Enter the name, ID, and optional description for the strategy.

    Strategy Display Name: A name for the strategy that will appear in the list of strategies in the administration interface. Each global strategy name must be unique. The name can be up to 50 characters long.

    Strategy ID: Commerce automatically creates a unique ID based on the display name you entered. You can change the ID while you are creating the strategy, but you cannot change it once the strategy is saved. An ID can contain only alphanumeric characters or numbers (a-z, A-Z, 0-9) and must be unique.

    Strategy Description: An optional description of the strategy that helps other users understand it. The description can be up to 1,000 characters long.

  4. Define the strategy by adding recommendation groups. By default, a blank recommendation group is already visible and ready for you to add conditions.
  5. Click the Add Condition button.
  6. Select a value from the Select a condition type list. You can select only one type per condition. For details about each condition type, see the table earlier in this section.
  7. If you selected Brand, Collection, Price, or Price Type, you must select an operator that links the property to a value that you will specify in the next step. The operators you see depend on the condition type you selected.
    • For Brand and Collection, you can select is, which matches the rule to the property you will select in the next step or is same as context, which matches the rule to .
    • For Price you can select from several numeric operators that match the rule to prices or price ranges. For example, select is less than if you want the condition to return all products priced at less than $50.
    • For Price Type you can select is, which matches the rule to a single price group, is one of, which matches the rule to more than one price group, or is same as context, which matches the rule to the shopper’s current price group.
  8. Select or enter a value for the condition property. For Collection and Price Type, Commerce displays a list for you to choose from. For Brand and Price, you must type in a value. The rule editor does not validate any text, numeric, or date values you type.
  9. To add another condition to the recommendation group, click the Add Condition button again and specify the details for the new condition.

    If a recommendation group contains more than one condition, it returns products that match all the conditions. Products that match only some of the conditions but not all, are not returned by the recommendation group.

  10. When you are finished adding conditions to the recommendation group, click Save.
  11. To add another recommendation group to the strategy, click Add Recommendation Group.
  12. When you are done adding recommendation groups, click Save to save the strategy.
  13. You must publish strategies in order for them to take effect on your production storefront, though once saved, they are immediately available for preview. See Publish Changes for more information.

Recommendation strategies examples

You could use the AI Recommendations rule builder to create the following sample recommendations strategies.

  • Suppose you want to create a dynamic cross-sell on the Product Detail Page that focuses on brand affinity. You could create a strategy with the following three recommendations groups. Set the maximum number of returned products in each group to four.
    • Group 1 includes two conditions: Brand is the same as context and Collection is not same as context.
    • Group 2 includes two conditions: Brand is the same as context and Purchased Together.
    • Group 3 includes two conditions: Brand is the same as context and Browsed Together.

    This strategy recommends up to four products that have the same brand but belong to a different collection, followed by up to four products that are the same brand and frequently purchased together, followed by the remaining products that are browsed together (but not necessarily purchased together).

  • Suppose you want to recommend an outfit when the shopper is viewing any apparel item. A custom recommendations strategy can dynamically generate an outfit recommendation that automatically promotes products from complementary collections. You could create a strategy with the following four recommendations groups. Set the maximum number of returned products in each group to one.
    • Group 1 includes two conditions: Collection is Men’s Tops and Collection is not same as context.
    • Group 2 includes two conditions: Collection is Men's Pants and Collection is not same as context, Number of Products = 1.
    • Group 3 includes two conditions: Collection is Men’s Shoes and Collection is not same as context.
    • Group 4 includes two conditions: Collection is Men’s Belts and Collection is not same as context.

    To display these recommendations, create a custom Product Details layout where you define the Product Type as Men’s Apparel and Men’s Footwear and place the Recommendations widget on the layout. Customize the widget to use this custom strategy by selecting it from the Strategy dropdown on the widget’s settings.

    Now, when the shopper is on any men’s apparel or footwear product details page, they will see recommendations from the complementary collections.

Create global exclusions

Global exclusion rules let you exclude certain products from all recommendations results, ensuring they are never recommended to shoppers. You can exclude products from recommendations based on their brand, parent collections, price range, or price type. Once you publish a global exclusion, it automatically takes affect on your site.

To select products to exclude from recommendations follow these steps:

  1. On the Marketing page, click Global Exclusions under AI Recommendations.
  2. Click the New Global Exclusion button.
  3. Enter the name, ID, and optional description for the rule.

    Global Exclusion Display Name: A name for the rule that will appear in the list of global exclusions in the administration interface. Each global exclusion rule name must be unique. The name can be up to 50 characters long.

    Global Exclusion ID: Commerce automatically creates a unique ID for the rule based on the display name you entered. You can change the ID while you are creating the rule, but you cannot change it once the rule is saved. An ID can contain only alphanumeric characters or numbers (a-z, A-Z, 0-9) and must be unique.

    Global Exclusion Description: An optional description of the rule that helps other users understand it. The description can be up to 1,000 characters long.

  4. Define the exclusion rule.

    Select a type of property to use from the Select a condition type list. You can select only one type per rule:

    • Brand lets you select from all the manufacturer values assigned to products in your catalog. For more information about the Brand property, see Create and edit product types.
    • Collection lets you select a collection whose products you want to exclude. You can select a collection from across all catalogs or from a specific catalog.
    • Price lets you select a price range of products to exclude.
    • Price Type lets you exclude products from specific price list groups.

    Select an operator that links the property to a value that you will specify in the next step. The operators you see depend on the condition type you selected.

    • For Brand and Collection, you can select is, which matches the rule to the property you will select in the next step or is same as context, which matches the rule to whatever product the shopper is currently viewing, as well as any products in the shopper’s cart or wish lists.
    • For Price you can select from several numeric operators that match the rule to prices or price ranges. For example, select is less than if you want to exclude all products priced at less than $50 from recommendations.
    • For Price Type you can select is, which matches the rule to a single price group, is one of, which matches the rule to more than one price group, or is same as context, which matches the rule to the shopper’s current price group.
  5. Select or enter a value for the condition property. For Collection and Price Type, Commerce displays a list for you to choose from. For Brand and Price, you must type in a value. The rule editor does not validate any text, numeric, or date values you type.
  6. Click the Create button when you are finished with rule.
  7. You must publish strategies in order for them to take effect on your production storefront, though once saved, they are immediately available for preview. See Publish Changes for more information.

Global exclusion examples

You could use the AI Recommendations rule builder to create the following global exclusions.

  • Never recommend anything from the brand MySample.

    In the Global Exclusion Definition, select Brand as the condition, select is as the operator, then MySample in the text box.

  • Never recommend anything in the Health and Wellness collection.

    In the Global Exclusion Definition, select Collection as the condition, select is as the operator, and then select Health and Wellness (including catalog context) from the Collection dropdown list.

  • Never recommend products that cost less than $25.

    In the Global Exclusion Definition, select Price as the condition, select less than as the operator, and then enter 25 in the text box.

  • Never recommend products that are on sale.

    In the Global Exclusion Definition, select Price Type as the condition, select is as the operator, and then select Sale Price from the dropdown list.