Create a Recommendation Strategy

Follow these steps to create a recommendation strategy.

Follow the steps in Create a Recommendation Strategy for Market-Level Recommendations if you want to use data from markets as part of a recommendation strategy.

Follow the steps in Create a Recommendation Strategy with Offline Purchases Data if you want to use data from an Offline Purchases dataset as part of a recommendation strategy.

  1. Click the Recommendations option from COMPONENTS in the top navigation bar, and then click Recommendation Strategies.

    Callout of the Recommendations Strategies tab

  2. Click CREATE RECOMMENDATION STRATEGY.

    Callout of the CREATE RECOMMENDATION STRATEGY button

  3. Select the option on the Recommendation Strategy Permission modal to make the strategy either global or local, and then click CONTINUE. For more information see Global and Local Recommendation Strategies.

    You cannot change the strategy permission after you click CONTINUE.

    Callout of the CONTINUE button on the Recommendation Strategy Permission modal

  4. Name the strategy. Click the placeholder title, type the name into the text field, and then click the green checkmark.
  5. If you're creating a local strategy and if the account has multiple product catalogs, then select one from Product Catalog.

    Callout of the Product Catalog selector

  6. Select an option from Recommendation Algorithm to determine which algorithm the strategy uses to populate the recommendations. For more information see Recommendation Algorithms.

    Callout of the Recommendation Algorithm selector

  7. If you selected Onboarded Recommendation Dataset in step 6, then select the dataset you want to use from Recommendation Dataset.
  8. If you selected a collaborative recommendation algorithm in step 6, then select from Base Recommendation on Items the type of customer behavior on which to base the recommendations.

    If you selected Onboarded Recommendation Dataset in step 6, then Base Recommendation on Items contains two additional options: No Lookup Key – Show All Items in Dataset and Customer ID. If you select one of these two options, skip step 10.

    Callout of the Base Recommendation on Items selector

  9. If you selected Item group ID(s) in custom variable in the previous step, then take the following actions.

    The option to base recommendations on item_group_id values passed at runtime through a custom variable is part of the Premium Recommendations package. Contact your dedicated Customer Success Manager (CSM) for more information.

    1. Type into Custom Variable a custom variable that your site passes to the Personalization platform using either the setCustomVariables method call in the Monetate API implementation or the monetate:context:CustomVariables in the Engine API implementation.

      The custom variable value can contain a comma-separated list of up to five item_group_id values.

    2. Optionally, select Pin products in custom variable to front of recommendation results if you want the products corresponding to the item_group_idvalue(s) derived from the custom variable to appear at the beginning of the recommendation results.

      If you select this option, be aware that pinned products configured in the recommendation strategy appear after any pinned products configured in a recommendations action that uses the recommendation strategy.

    Callout of the 'Custom Variable' field and the 'Pin products in custom variable to front of recommendation results' option

  10. If you selected a collaborative recommendation algorithm in step 6 and in step 8 selected Viewed, Carted, or Purchased, then select the session scope option from From on which you want to target the customer behavior. The options available in From are determined by the recommendation basis you selected in the previous step.

    The Last Carted Item(s) option only considers the very last product put into the cart. It does not consider multiple products.

    The page referenced by the Item(s) on the Page and First Item on the Page options is the page on which the recommendation action exists. All other options reference the behavior of the customer currently exposed to the recommendation action.

    Callout of the From selector

  11. Select an option from Lookback Period if you selected a recommendation algorithm that requires a time frame from which to collect data for determining recommended products.
  12. If you selected an eligible recommendation algorithm in step 6, then select an option from Geographic Targeting if you want the strategy to also consider the customer's location to populate the recommendations:
    • Country targeting — Only products relevant to the customer's country are recommended
    • Region targeting — Only products relevant to the customer's region, as defined in MaxMind's GeoIP2 database, are recommended

    Callout of the Geographic Targeting selector

  13. Optionally, toggle Randomize Results to YES if you want the order in which recommended products appear in the slider to be less systematized.

    Callout of the Randomize Results toggle

  14. To further refine the items included in the strategy, click ADD FILTER, select an option from SELECT ATTRIBUTE, and then complete the filter equation. Repeat this step as necessary to add as many recommendation filters as you believe the strategy needs. For more information see Filters in Recommendation Strategies.

    Callout of the ADD FILTER button

  15. Optionally, configure up to five Boost and Bury filters to influence if recommended products that meet that filtering criteria are more likely (boost) or less likely (bury) to appear for the customer.

    Contact your dedicated Customer Success Manager (CSM) if you want the Boost and Bury feature enabled.

    1. Click ADD ATTRIBUTE.

      Callout of the ADD ATTRIBUTE button for the Boost and Bury feature

    2. Select an option from SELECT ATTRIBUTE, and then complete the filtering equation.

      Callout of the attribute-based filter equation

    3. Select Boost to promote the products that meet the filtering criteria, or select Bury to suppress them.

      Callout of the selector to boost or bury recommended products that meet the filtering criterion

    4. Adjust the slider to determine by what percentage the products that meet the filtering criteria are boosted or buried.

      You can only set the percentage using the slider and cannot type a number into the text field to the left it. Furthermore, you can only adjust the percentage in increments of 10.

      Callout of the percentage slider

    5. Repeat steps 14a through 14d to add up to four more independent Boost and Bury filters. See Using Multiple Boost and Bury Filters to better understand how having more than one Boost and Bury filter can impact the recommendations.

      Example of three Boost and Bury filters created for a recommendation strategy

  16. Click SAVE.

    Callout of the SAVE button

After you save the strategy, you can preview it from the configuration page in certain situations. See Preview a Recommendation Strategy for more information.

Using Multiple Boost and Bury Filters

While Boost and Bury filters look much like recommendation filters, they have some key differences:

  • You can only add a total of five Boost and Bury filters to a recommendation strategy.
  • You don't join multiple Boost and Bury filters with the AND or the OR logical operator.
  • The order in which you add multiple Boost and Bury filters to a recommendation strategy doesn't impact how much a qualifying product is ultimately boosted or buried because each filter is independent and not connected in a logic sequence with the filter that might be listed before or after it.

To better understand how Boost and Bury filters impact a product's final relevancy score for a recommendations action, consider the following example.

A recommendation strategy uses the Top Selling by Purchase Count recommendation algorithm that considers 30 days of historical data and doesn't include geographic targeting. It has a recommendation filter based on the age_group attribute to exclude products for infants and children. At this stage of processing, five products identified by the recommendation algorithm and recommendation filter have these relevancy scores:

  • Product A: 50%
  • Product B: 30%
  • Product C: 20%
  • Product D: 10%
  • Product E: 5%

The recommendation strategy's three Boost and Bury filters are then applied:

  • Quantity < less than "1000" BURY –90%
  • Product type = equals (Starts With) "Jackets" BOOST 50%
  • Price ≥ greater than or equal to "59.95" BOOST 80%

Of the identified five products, only products B and C meet at least one of the Boost and Bury criteria:

  • Product B
    • Product type = "Jackets"
    • Price = "$49.95"
    • Inventory = "100"
  • Product C
    • Product type = "Jackets"
    • Price = "$79.95"
    • Inventory = "2000"

Two of the three Boost and Bury filters apply to product B, and the impact of those applicable filters to product B's relevancy score is calculated as follows:

30% × (1 – 90%) × (1 + 50%) = 4%

Like product B, only two of the Boost and Bury filters apply to product C, but both filters are boosting measures. The impact of those filters to product C's relevancy score is calculated as follows:

20% × (1 + 50%) × (1 + 80%) = 54%

Because of the impact of the Boost and Bury filters on just these two products, the final relevancy scores for the five products are as follows:

  • Product C: 54%
  • Product A: 50%
  • Product D: 10%
  • Product E: 5%
  • Product B: 4%

The impact of the Boost and Bury filters on the final recommendations relevancy scores of products B and C wouldn't change if those filters were listed in a different order within the recommendation strategy because of how the calculations are made.