Product recommendations (results) are delivered to locations (pods) across your retail site. Product recommendations (results) are generated and ranked by our AI algorithms (strategies). Constructor's recommendation service delivers personalized, attractive recommendations in a wide range of contexts. Constructor uses past user search, browse, and purchase behavior to display highly customized recommendations, individualized to each and every user.

• Personalized
  Constructor informs recommendations by everything known about a particular user, not just the products that users buy with a particular item.
• Collaborative
  Constructor looks across all customer histories to identify common patterns, assembling a dense graph of connections based on item-item and user-user relationships.
• Conversion-optimized
  Recommendations are constantly validated against observed click and conversion behavior.
• Integrated
  Recommendations use the exact same personalization signals and graph algorithms as our search engine to provide a cohesive experience.
• Product inheritance
  For new or less popular products, Constructor marries behavioral signals with content embeddings by mapping new or less popular products to the most similar products with view and purchase information using titles, categories, descriptions, keywords and other product data.

Product recommendations

Recommendations return lists of ranked products that are most attractive to drive specific business KPIs and help move shoppers along their shopper journey. The most common image of recommendation results that comes to mind for most shoppers is a set of bundled product recommendations directly underneath the main product image on a product detail page (PDP).

Recommendations pods

Recommendations pods can be served in specific pages and page locations within the retail site.

Examples of site locations are

• The Home Page
• A Product Landing Page (PLP)
• A Product Detail Page (PDP)
• The Cart page

Examples of locations within those pages include

• Top of the page
• Within results
• In popups or modals
• At the bottom of the page.

Every location you would want to serve recommendation results should be thought of as a recommendations pod. Pods represent a specific location on a page or set of pages (template) where recommendations may be rendered. A page may have one or more pods.

Recommendation strategies

Strategies represent a particular algorithm used to generate recommendations results. Currently, a pod can be assigned a single strategy, but the capability is possible in the future to support multiple strategies in a single pod. Example strategies might be "Recently viewed items," "Alternative items," etc. Constructor provides several strategies that can be associated with a given pod.

The strategies available for merchandisers to configure on the dashboard are documented below.

Alternative

By collecting and analyzing co-occurrence data based on user behavior, Constructor can recommend similar items. The result will show products a user might consider as an alternative to a particular product or set of products. For example, if a user is looking at toothpaste, this algorithm shows other types of toothpaste.

Additional customizations can be made to the strategy by adding unique conditions, specific to different business needs. Conditions allow for hiding or slotting items or attribute-based item groups.

Bestsellers

Shows most popular items for a specific time frame. By default, bestsellers will show the most popular items over the last 14 days. This detail can be customized to show bestsellers over the last 7, 14, or 30 days. Contact your Customer Success Manager if you wish to implement this customization.

Complementary

Constructor evaluates co-purchase behavior and other conversion signals to dynamically create lists of products that users often buy together with a given item. These are often distinctly different from alternative recommendations, and are useful to use in conjunction with other suggestions or in a different context (such as a checkout page). This results in showing products a user would purchase in addition to a particular product or set of products. For example, if a user is looking at toothpaste, this algorithm shows toothbrushes and dental floss.

Just like the alternative strategy, additional customizations can be made to the strategy by adding unique conditions, specific to different business needs. Conditions allow for hiding or slotting items or attribute-based item groups.

Bundles

Bundles are an advanced complementary strategy that return items which are most frequently purchased in sets together. The bundle strategy optimizes for products that are likely to be added to cart as a whole set with the primary product on a PDP. Please reach out to your customer success manager to learn more about best practices for implementing the bundle strategy.

Filtered items

A pod with this strategy will recommend items matching a provided filter expression. An example would be if you sell educational books to an audience of teachers, this strategy could be applied on product pages to show products that are within the same specific grade level of that product. Within the parameters of the filter being applied, the most popular items will be displayed.

The requirement for this strategy is a expression in the specific format that would be provided in browse or search endpoints. Several facets can be included. Specific items can also be prioritized by providing an optional item ID.

More detail about enabling and customizing the filtered items strategy can be found in our API reference documentation

Query recommendations

Query recommendations recommend items based on what users have clicked, added to cart, and/or purchased after a zero-result query. This pod is intended specifically for pages that result in zero results.

Recently viewed

The last items the user has clicked on will be shown when utilizing this strategy. This applies to items that were last seen while either searching or browsing.

User featured

This strategy focuses on items the unique customer is likely to buy. The items displayed are based on the specific user's prior behavior and are ranked highest to lowest based on this personalization score.

Additional features

Remove Converted Items - By default, this feature is turned off. This means, items converted on by the customer could still show up in a recommendation pod. This feature could be turned off if it does not align with the needs of your business. Contact your Customer Success Manager if you wish to implement this customization.

Filter out items with same naming convention - For customers with products in multiple variations, whose items are recognized as unique products, those products with the same name can be filtered to not repeat within a pod. For example, an apparel company with a product ‘corduroy pants’ that comes in multiple colors can choose to filter out items with that same naming convention (“corduroy pants”) from appearing multiple times in the recommendation pod. This feature is not activated by default but could be activated by reaching out to your Customer Success Manager.

Not Enough Items Default - If there are not enough items that meet the criteria for a particular recommendation pod, Constructor will automatically plug in another strategy to fill the remaining spots. By default, Constructor will use the Bestsellers strategy in this scenario. This logic can be turned off by request.

Personalization - By default, personalization is added on all recommendations. This feature could be turned off by request, with the exception of the ‘User Featured’ strategy as that is based entirely on personalization.