Product recommendations are a significant addition to your e-commerce website. They contribute to:

• Boost in conversions;

• Increase in shopping cart value;

• Improvement of website UX;

• Rise in return visits.

Types of Recommendation Systems

There are three types of recommendation systems:

• Collaborative filtering;

• Content-based filtering;

• Hybrid system.

Collaborative filtering

Such systems make recommendations based on the other users’ history of actions (click, add to favorites, purchase, review, etc.). Recommendation algorithms come in handy when your store has a lot of traffic because to make a prediction it needs to see how different website visitors behave. It analyzes the set of users’ preferences and actions, finds similarities between them, and arranges users with the same patterns into groups. If a new customer behaves the same way as a group of users in the past, it will predict that he or she will have an interest in the same products.

For instance, say Shopper X is browsing a MacBook. Then he opens a search box and types “smartphone.” Based on the previous users’ history of actions, the engine can guess that the smartphone this customer is looking for is probably an iPhone. So, the recommendation system shows it to him first.

Pros:

• Works when product data is insufficient – can generate recommendations even if the item has a small number of attributes, mistakes, or missing information.

• Becomes more accurate as it learns more about users.

• Can make suggestions across different product types.

Cons:

• “Cold-start problem” – for the recommendation system to start working properly, it needs data on how the user bought and rated products.

• If the product is new, and no customer has yet bought or reviewed it, the recommendation system won’t show it.

• Superseded products – the system has data on a previous version of the product, but not the new one, and even though the versions are pretty similar, the engine will view it as two completely different products. So, the system needs time for users to rate the new version before it can recommend it to website visitors.

• Requires history on a large number of customers to work more precisely.

Content-based filtering

A content-based filtering recommendation system analyzes the content that interests the user and suggests products with similar properties or characteristics. The engine analyzes the item profile (the set of product attributes) and user profile. It looks at how previous customers rated it and reviewed it, and whether they bought the item or not. Then it estimates what product the current user is browsing for and suggest products with a similar set of attributes.

For example, say User Y browses for “Nikon Z 6 Mirrorless Digital Camera.” The recommendation engine will suggest items with similar attributes like ‘’24-70mm f/4 S zoom lens,” “FTZ Mount Adapter,” and “72mm UV filter.” These items will fit the camera and will catch the attention of potential customers.

Pros:

• No “cold-start problem” – recommendations can be given as soon as the system is implemented.

• Transparent - easy to explain the basis of recommendation.

• Easy to implement.

Cons:

• The danger of a “filter-bubble” emerging, when the system recommends products only within one filter option, even if it is reasonable to suggest products from different categories.

• Needs well-structured product data (accurate attributes, no missing or inconsistent data).

• Unlike collaborative filtering, requires human intervention to configure recommendation rules.

• With time, collaborative filtering might become more precise than content-based filtering.

Hybrid systems

This type of recommendation system helps to provide personalized recommendations to its users. Hybrid systems combine collaborative filtering and search-based filtering approaches as the two methods have long been viewed as complementary.

The Convermax Recommendation System in Action

We use a content-based method to provide product recommendations and apply Machine Learning techniques to rank products. They start working from day one of the search engine's implementation.

The recommendation widget is set on the product page. An e-commerce merchant can customize the exact location of “You May Also Like” suggestions on the page.

The engine analyzes the attributes of the product the customer is viewing at the moment and uses the set of them to create a search results page with products that fit it. The order of the products depends on the algorithm that estimates their ranking. Our recommendation system takes into account product trends, customer behavior logs, and buyer profiles to define product ranking and to make suggestions. If an e-commerce store has data on offline sales or history of sales on your previous e-commerce website, we can load it, make the engine analyze it, and provide better recommendations from the start. And as we gather the data, the search engine analyzes it and becomes smarter.

Also, if you want to promote some specific products you think will interest the customer, we can allocate them manually and mix them with products that are suggested by the engine.  New and best-rated products are automatically prioritized in search results and recommendations.

For instance, Incredible Rugs and Décor has different types of recommendations: one that suggests similar products based on the colors of the products, and another that shows products based on previous customers’ behavior.

You can click on the color pallet of the product to get to a page with rugs of similar colors.

“You May Also Like” recommendations in this store are formed based on product colors and other selected attributes, number of views, users’ reviews, sales data, and rating.

Usually, the recommendation widget on the product page displays a few products and has a “Browse all similar products” button which will redirect customers to the search results page with more product suggestions.

Want to try the Convermax recommendation engine on your website? This feature comes at no extra cost. Schedule a call with us to learn more.