What Are Recommendation Algorithms

UI Path: New Campaign > Recommendations > Any Recommendation Wizard

Recommendation algorithms take everything that Personyze knows about your products, your visitors, and the interactions between the two, and uses that data to automatically present each visitor what they are most likely to be interested in seeing, at any given time, on any given page, or in emails.

When recommendations are presented, whether in horizontal or vertical arrays, as sliders, or even banners, we call these Recommendation Widgets. Each widget is a unique recommendation, with its own algorithm.

You can and probably will have various widgets with various algorithms all over your site, and every time a visitor loads a page with one or more of these widgets, the products they recommend will be slightly different, because Personyze is constantly and automatically learning from how visitors interact with the products or content on your site.

The Key to Understanding Recommendation Algorithms

The key concept to understand, when it comes to recommendation algorithms, is that they are a hybrid system which combines crowd data (What EVERYONE is doing) with individual interest data (what THIS visitor is doing).

For instance, you may have a recommendation algorithm which shows :

“Most Viewed products (crowd data), based on visitor’s most recent interest (individual interest data).”

This will show the visitor items from the category they most recently showed interest in, and will prioritize those items according to how often they’ve been viewed by visitors as a whole.

There are a few exceptions to this, however:

• Buy It Again Recommendations are not actually an algorithm, per se, they are simply a list of items which the visitor bought previously
• Abandoned Cart Recommendations are simply a list of items the visitor left in their cart.
• The Auto-pilot Recommendation utilizes more in-depth AI to find patterns in your visitor’s behaviors, examining various factors such as location, gender, and other variables. Using this, it will show what the visitor is most likely to buy, based on a variety of factors beyond just known interest. This algorithm, however, does require a certain amount of interaction data before becoming effective.

Still, most recommendation algorithms involve this hybrid approach.

Filters

You can also add filters to any given recommendation, to make it more specific. These filters will either be applied to the crowd data (population whose data we are using) or the product.

For instance, on the algorithm from our previous example, most viewed based on recent interest, we could add the filter “Age Range”, which would show the most viewed among the visitor’s age range, rather than among all visitors, as it would be otherwise.

On the other hand, if you wanted to add a product-based filter, you might add the “New in Stock” filter, to show the most viewed new in-stock items from the visitor’s most recent interest.

Best Practices

With this information in mind, you should be able to determine which algorithm makes the most sense for the location on the site (or in emails) where your widget is being presented. Here are some general guidelines, to help you get started:

1. On the Homepage – You will usually want to show more general recommendations, such as:
   “Most Viewed Products in all time from visitors Any Interest”
2. On Product Pages – You will want to show widgets that are more specific, such as:
   “Most Bought Products in last week from visitor’s Most Recent Interest”
   (most recent interest category, when placed on a product page, will always be the category of the current product.
3. On Cart Checkout Page – You may want to show the visitor items which have a cross-sale association in your catalog with the items in their cart, such as:
   “Most Bought cross-sale products in last week from visitor’s most recent purchase/add to cart interaction.”