Recommendations: Help users reach the goals you've set for them

After you identify a predictive goal for your users, create the recommendations most likely to help users reach that goal. Amplitude’s AutoML determines which items are most likely to maximize each user’s predictive goal, and then places those items in front of the user.

Amplitude Activation's machine learning algorithm clusters your selected users into groups of similar users. Shared user properties and past behaviors determine this similarity. Next, Amplitude analyzes historical data to identify which items are most likely to increase each cluster’s propensity to convert. Finally, Amplitude assigns a ranked list of items to each user based on their assigned cluster.

The algorithm retrains every hour, so it incorporates new property and behavior information into its results.

Best fit for recommendations

Amplitude Activation works best for user-based personalization, not account-based personalization. Recommendations are most useful for companies that need to show an array of items, such as products, articles, or shows, in a product carousel, product list, or cart flow. Ecommerce, marketplace, B2C, and subscription software companies are the best fit for Amplitude Activation.

Enterprise B2B companies, on the other hand, are unlikely to benefit from using recommendations.

Recommended use cases

Amplitude Activation is not an analytics feature. It’s a personalization feature that helps you improve in-product and digital experiences to maximize lift. Recommendations work best for three user-based digital commerce personalization types: assortment, next-best action, and cross-sell.

• Assortment: Ranks items to display on a homepage or within a category page. These items can be SKUs, articles, shows, and similar content. Assortments are appropriate for increasing engagement.
• Next-best action: This scheme identifies a second item the user might be interested in and places it into the checkout or carousel flow, or in an email after purchase. Here, the objective is to increase conversions.
• Cross-sell: Ranks items that signify lifecycle stages the user hasn’t yet achieved. These items are usually categories, products, or subscription types. The primary objective is to increase LTV.

Support for other use cases, such as in-session recommendations and new item recommendations, is in development.

Data requirements for a recommendation

There are three data components to configuring a recommendation: the outcome event, the exposure event, and the event property. For recommendations to work, you must instrument the data behind these components in your taxonomy:

• The outcome event is the objective goal for your recommendation. Often, the outcome event is something like “purchase” or “subscribe.” Track this outcome as an event in Amplitude Analytics.
• The exposure event is an action the user takes before the outcome event. Typical exposure events include “add to cart,” “click product,” or another event with an event property that configures an item for recommendation. Track this event upstream in the conversion funnel of the outcome event.
• The event property is the “item” that appears in the recommendation. This property often has a name like “SKU,” “ID,” “name,” “category,” or “brand.” Store this information as an event property on the exposure event.

Work closely with your Amplitude CSM to ensure these conditions are met.

To learn more, read Build a recommendation and Use recommendations in personalization campaigns.