How to build recommendations (recommendation builder variant)

The AB Tasty Recommendation Builder allows you to create tailored lists of items to recommend, using a combination of algorithms and transformations. This flexible system lets you define precisely how recommendations are generated and modified to meet your business needs.

Recommendation Structure:

A recommendation is composed of:

• Algorithms: Generate a list of items.

• Transformations: Modify the list generated by an algorithm.

Drag-and-Drop Interface:

• Build your recommendation by dragging and dropping operations into drop zones.

Drop Zones

FOLLOWED BY Divider:

• Concatenates results from multiple algorithms, avoiding duplicates.

Example: Combine products from two algorithms (e.g., "Most Popular" FOLLOWED BY "New Arrivals").

THEN Divider:

• Applies transformations to modify algorithm results.

Example: Filter, shuffle, or sort the list.

THEN for Transformations:

• Allows multiple transformations in sequence, separated by THEN.

Example Recommendation Scenario: Create a Multi-Source Recommendation Algorithms:

First 15 products from [ALGO] User Reco. FOLLOWED BY 15 products from [ALGO] Most Popular. Filters Applied to All Products:

Recommendable in newsletter = True. Marketplace = False. Accessory = False. Pickup in store = False.

Steps to Build a Recommendation

Add an Algorithm:

• Select an algorithm and configure its settings.

Transform Algorithm Results:

• Drop a transformation into the algorithm box (e.g., filter or shuffle).

Add a FOLLOWED BY Divider:

• Drop a second algorithm after FOLLOWED BY to concatenate results (e.g., as a fallback algorithm).

Apply Global Transformations:

• Add transformations after the THEN divider to modify the overall results.

Add Exceptions:

Use the "+ Add Exception" button to apply alternate algorithms based on specific conditions.

Preview Your Recommendation

Once your recommendation is built:

• Preview Button: Click to view results.

• Modal Options:

  • Set parameters.

  • Preview results as a table or list.

  • Inspect results for each algorithm.

  • Display additional details for each item.

  • Check API calls leading to the result.

Operations

Algorithms

• Sorted Items: Example: Top 12 items by revenue over the last 30 days.

• Associated Items: Example: Items frequently purchased with the selected item.

• Similar Items: Example: Items often viewed with the selected item.

• Recommended Items: Reuse a saved recommendation.

• Handpicked Items: Add items manually or by importing item IDs.

• Used Items: Example: Last 12 items bought by a user (requires user data integration).

Transformations

Filter:

Include only items matching a condition. Example: Show only items where the brand is "Apple".

Dynamic Filter:

Use input variables to make filters dynamic. Example: Keep products cheaper than the input item.

Sort:

Sort items by a specific field. Example: Sort by top sales.

If Condition:

Apply transformations conditionally. Example: If brand = "Apple", show Apple items; if brand = "Dyson", show Dyson items.

Shuffle:

Randomise the order of items. Use cautiously, as it can disrupt relevance.

Limit:

Restrict the number of items displayed. Example: Limit results to 20 items.

Exclude:

Remove items based on input variables. Example: Exclude items already purchased by the user.

Final Notes

• The Recommendation Builder offers powerful tools for creating highly customised recommendations.

• Regularly test and preview your configurations to ensure they meet your goals.

• For additional guidance, consult the platform documentation or contact your Customer Success Manager (CSM).