Measuring impact of recommendations - Amazon Personalize

Measuring impact of recommendations

As your customers interact with recommendations, you can measure how the recommendations are helping you achieve your goals. You can identify which campaigns and recommenders have the most impact, such as which resource generates the most minutes watched or the most clicks. And you can compare the performance of Amazon Personalize recommendations to those generated by third-party services.

To measure the impact of recommendations, you create a metric attribution. A metric attribution creates reports based on the interactions and items data that you import, and the metrics that you specify. For example, the total length of movies watched by users, or the total number of click events. Amazon Personalize aggregates calculations over a 15-minute window. For PutEvents and incremental bulk data, Amazon Personalize automatically sends metric reports to Amazon CloudWatch. For bulk data, you can choose to publish reports to an Amazon S3 bucket.

For each interaction that you import, include source data to compare different campaigns, recommenders, and third parties. You can include the recommendation ID of the recommendations you showed the user or the event source, such as a third party.

For example, you might have a video streaming app that shows movie recommendations from two different Amazon Personalize recommenders. If you wanted to see which recommender generates the most watch events, you could create a metric attribution that tracks the total number of watch events. Then you could record watch events as users interact with recommendations, and include the recommendationId in each event. Amazon Personalize uses the recommendationId to identify each recommender. As you record events, you can view the watch event totals aggregated over every 15 minutes for both recommenders in CloudWatch. For code samples that show how to include a recommendationId or an eventAttributionSource for an event, see Event metrics and attribution reports.