How Do eCommerce Product Recommendation Engines Work?

While it’s possible to manually implement rudimentary “also-liked” recommendations on your ecommerce site, product recommendations best practices call for the deployment of a ‘product recommendations engine’.

There are three basic approaches used to configure the underlying algorithm:

  • The content-based filtering method analyzes customer data on the likes and dislikes of each user (cookies allow tracking over multiple visits), then makes recommendations based on the browsing history of that user. The idea behind content-based filtering is that if you enjoy a certain item, you’ll likely also enjoy a similar item. An example of a content-based filtering system would be if you were listening to Pandora and consistently ‘liked’ downtempo jazz music. The filtering system would take that information and begin recommending similar music to you based on the songs you preferred. 

  • The collaborative-filtering method incorporates data from users who have purchased similar products, then combines that information to make decisions about recommendations. The advantage to this filtering method is that it is capable of making complex recommendations on items such as music or movies without having to ‘understand’ what the item is. This method of filtering operates under the assumption that users will prefer recommendations that are based on purchases they made in the past. Here’s an example: If customer A likes a specific line of products that customer B also likes (assuming they have similar interests), then collaborate-filtering would assume that the customer A would like other products that customer B purchased and vice versa. 

  • A hybrid method combines the content-based and collaborative-based methods to incorporate group decisions, but focus the output based on attributes of a specific visitor. An example of a hybrid filtering system would be how Spotify curates their personalized ‘Discover Weekly’ playlists. If you’ve ever listened to a personalized Spotify playlist, it’s shocking how accurately they’re able to recommend songs based on what you like. The secret behind how they pull this off is through a complex hybrid filtering system that aggregates data on your listening habits as well as similar users’ listening habits, to create a playlist of unique songs that align with your personal taste. 

All three methods use machine-learning algorithms to fuel the process and provide personalized product recommendations. While the mathematical principles behind each are elaborate and complicated, the application to your online store doesn’t have to be overwhelming. 

What Are the Benefits of a Product Recommendations Engine?

Is the product recommendations process really worth the trouble?

Isn’t the incorporation of machine learning a bit beyond the scope of all but the largest ecommerce websites?

Those are the types of questions we often hear from clients. There are times when it seems the high-tech movement is going too far, and machine-learning algorithms are a prime example of that complaint.

Given the potential benefits, though, the argument often settles itself. When a tool proves itself sufficiently valuable, the question moves from “Why?” to “How?”

  • Research conducted by Barilliance in 2018 concluded that product recommendations accounted for up to 31 percent of ecommerce revenues. On average, customers saw 12 percent of their overall purchases coming from products that were recommended to them. 

  • A Salesforce study of product recommendations concluded that visits where the shopper clicked a recommendation comprise just 7 percent of total site traffic, but make up 24 percent of orders and 26 percent of revenue. 

  • The conversion rate for visitors clicking on product recommendations was found to be 5.5x higher than for visitors who didn’t click

  • A Gartner study predicts engines that gauge and react to customer intent will be capable of boosting ecommerce profits as much as 15 percent by 2020

  • As online shoppers become more used to personalization, they equate it with professionalism – meaning your site needs to bump up to keep up

  • An Accenture report says personalization increases the likelihood of a prospect purchasing from you by 75 percent

Studies increasingly show the value of product recommendations and the critical role they play in personalization strategies. Recommendations not only lift conversion rates, they help deliver improved user experience to keep visitors coming back and can boost the average order value. 

Once an ecommerce manager is convinced of the benefits of a product recommendation engine, the next step is to determine product recommendation best practices and configure the product recommendation algorithm accordingly.