Today, we’d like to discuss our product recommendation service features:
- Using product recommendations: where and how
- Mindbox algorithms
- Setting up automatic recommendations
- Future plans
Where and how to use product recommendations
There are several ways to generate recommendations in our product:
- use the Mindbox algorithms:popular products,similar products, and related products
- upload your recommendations in a file
- set up a third-party service integration (RichRelevance, Rees46).
Recommendations in messages
Our clients use this module in manual and trigger-based messages. The most popular campaigns are:
- welcome sequence
- abandoned product view
- abandoned category view
- abandoned cart
- product is (not) available
All of the Mindbox algorithms are available for email messages.
Recommendations on your site
All of the Mindbox algorithms are available for authorised clients. Anonymous users can access the «Popular products», «Similar products» and «Often bought together» algorithms.

Example from AMwine.ru
Mindbox algorithms
The minimum that you need to generate recommendations using our algorithms is a live tracker. A tracker is a javascript code that collects customer activity data from your site. We also need order history data for the «often bought together» algorithm.
You can further improve your results from some algorithms («often bought together» and «similar products») by uploading more information about your products. For example, we were able to significantly improve the quality of recommendations at MirKrestikom.ru by uploading additional information about their products: subject information and embroidery technique.
Our algorithms provide transparent results: we can always explain why product «A» recommended product «B» and, if needed, we can fine-tune the display properties.
«We’ve been testing Mindbox’s new recommendation algorithm. It looks cool! Our customers see product recommendations based on their interests. Shape, color, size – almost anything! As if it’s not an automatic letter, but rather someone sat down and selected products to please the customer.»
Tatiana Kuznetsova, Project Manager at Divan.ru
1. Popular products
What we do: We choose the most popular products based on views or orders.
This is a universal algorithm that works equally well on a website’s homepage as in the product card, as well as in manual and trigger-based messages.


2. Popular products from a related category (optional: and with similar prices)
What we do: We chose popular products from categories with goods that the user has interacted with.
In emails, this could be products that they’ve viewed, left in their cart, or purchased; on your site, it could be product cards or categories that they’ve viewed.
The algorithm is based on the original product’s category and works best for:
Messages:
- abandoned product view
- abandoned category view
- abandoned cart
- product is unavailable or has become available
Website:
- adjustable category page
- product card
3. Similar products
What we do: We choose the products that are most similar to the original product based on several criteria (manufacturer, price, or individual characteristics).
The criteria can be prioritised to select products that are most similar to the qualities that you think matter the most.
This works well for triggers:
- abandoned product view
- abandoned cart
- product is unavailable or has become available
On websites, it fits nicely in the product card.


4. Often bought together
What we do: We choose the products that are most often bought with the original product.
The best use cases in messages:
- abandoned cart
- product has become available
It looks good on your website’s product card.


You can combine as many algorithms as you want:

Sample message from Hoff: a combination of the «Popular products in this category» and «Related products» algorithms.
Setting up automatic recommendations
You can set up automatic recommendations based on the number of views or orders by choosing a time frame and the number of products to randomly select. To avoid showing the same products, the specified number of recommended products is selected each time.

Message settings can be based on the user’s last session, as well as on information about their most recent product views or purchases.


Creating the operations for an product recommendation service for use on a website
Future plans
Our next steps include developing:
- segment management interface for website recommendations
- A/B test function interface for website recommendations
- new personalisation algorithms
More to come 😉