Recommendations generate an important part of the revenue among the biggest players in the world such as Amazon, Netflix or Ebay. Product recommendations by Bloomreach Engagement could be used in various placements:
- Homepage: To show the webshop sells awesome products (or to reactivate, personalise, etc.).
- Category detail: To show the most popular items within the category that is currently viewed by a user.
- Product detail: To show alternatives for a currently viewed product by a user.
- Basket: To suggest other relevant products to these already in the basket.
- Email campaigns: To personalize email marketing strategies or just to show which products the given webshop is selling.
- Web layers, push notifications...
In general, Bloomreach Engagement Recommendations leverage all customer and catalog data to find the most suitable items for each individual customer in a given situation. Due to the power of artificial intelligence, Bloomreach Engagement recommendations enhance user experience through advanced personalization resulting in an uplift across multiple KPIs.
Watch these two short introductory videos about this recommendations:
By providing multiple pre-built recommendation engine templates suitable for a variety of different placements and use-cases, you can get all the benefits of AI in a few clicks. Bloomreach Engagement recommendations are not a black-box - it is possible to combine different engines and build a solution that fits precisely your needs. Most importantly, as in the whole Bloomreach Engagement platform, everything is done in real-time.
To create a new RS in Bloomreach Engagement, you first need to build an engine using pre-built templates in the Bloomreach Engagement. Go to
+ button in the top-right corner and pick a suitable template for your use-case. See the picture below for visual guidance. Now you can create a new recommender system.
Each template is discussed in the next chapters with examples of ideal setup. You can choose from the following list:
In Bloomreach Engagement we recognize two more distinctions between various RS:
1. How the recommendation model is created:
Recommendation models are divided into two groups:
- Rule-based model - a family of methods based on some simple heuristics (e.g. pure statistical approaches). This computation is very quick and the model is ready to be used immediately.
Example: top selling products, most viewed products, etc.
- AI model - computed by a machine learning algorithm based on the historical data (the process of learning is often called “training”). Each training of an AI model takes several hours.
Example: TF-IDF, Matrix factorization
2. How we request models:
There are two main types of recommendations requests:
- Recommendations based on their past events.
- Recommendations based on the context of the customer's last actions regarding the product they are currently viewing. There was not enough time to load these actions as events, so they are taken into account by the recommendation through Recommendations for customer API call. An example of such actions would be viewing a product or adding it to the cart.
Example: Recommending alternative products for a product being currently viewed on the product detail page.
A catalog must be imported in order to create recommendations. It is necessary to make sure:
- To drag item_id label to the column which indicates unique product id and match the same id as defined in the events such as view item or purchase item.
- To index all of the columns which will be used in the recommendations catalog filter (be aware that indexing is not possible to change after you import the catalog)
- To set labels for columns image, URL, title, price in order to see recommendations results in the single customer view and test tab of each recommender engine.
*Catalogs could be found in the Data & Assets -> Catalogs.
Most of the recommendation engines require event tracking in order to understand and make use of customer behavior. Make sure that event tracking is set up correctly, especially in the case of:
View_item which serves to track what items customers are viewing
Cart_update which tracks what items are added or removed from cart
Purchase_item which tracks what items were purchased
Data tracking refference
For more information on the even tracking see Data Manager
In order for all of the recommender systems to work correctly, please fill the following events in the part Data & Assets -> Data manager ->Mapping: Purchase item, Add to cart, View item, View category.
Updated 3 months ago