How to build product sets based on AI models

8 min read
product sets based on AI models cover photo

One of the interesting solutions that is used to increase the value of the basket is to offer the customer product sets. However, they cannot be random products, so their selection must be well thought out. AI algorithms can help us in this and allow us to easily create effective sets of products that not only affect conversions but also look good on our website.

Main assumptions 

Why can product sets be so engaging? 

A product set is basically a set of products from the same or different categories which can be presented together and be purchased as a group. They can be set up manually, but it is more effective to use AI models to create them automatically and make them more engaging. 

How to set it up 

Step 1 CHOOSING THE RECOMMENDATION TYPE

The main challenge is to automatically create relevant product sets for any product based on AI models.

First of all, you should remember that It will be not enough to set only one campaign for all products in your stock. We often need to create separate models and separate campaigns for the various categories of products. Which types of recommendations are the best for product sets? 

  • Similar recommendations (similarity based on attributes such as a name, description, category etc.)
  • Complementary recommendations (finds products that are generally bought together).

Step 2 PRODUCT SETS MODEL CONFIGURATION  

After choosing the AI model, we need to configure its settings properly:

  • At first we define the number of products that will be returned in our set campaign.
  • After that we can optionally define filters, such as a category, price, brand, discount, gender or other custom attributes. Filters are the basis of a good product set. If we set it properly, then you will be satisfied with the effects. 

We need to pay attention to a few additional settings regarding filters. It's very useful to use boosting based on our metrics. It allows you to increase the priority of some products which are, for example, more commonly bought, viewed or added to the basket.  

It is also important to remember that not only can we include the category or custom attributes, but we can also exclude them. It's very helpful especially if someone wants to present recommended products from any category but for the one to which the main product belongs to. 

STEP 3 TESTING

When we prepare an AI model properly, we can test it and check to see if it delivers satisfying results. For each created model, we can make a preview and put in the preview any product to which you will generate a product set.

Example of prepare an AI model properly and test

STEP 4 PLACING ON THE WEBSITE  

When the model is ready, we need to finally place it on the Website. We can do it in many ways e.g. using our API, but also using our dynamic content. It is a necessary step because AI model itself will not be able to display the product set on the website straight away. You need to prepare an HTML template and put it in the proper place on the product page. Dynamic content will be the easiest way to do it as it allows you to use easy-to-implement inserts from AI campaigns. 

Main Challenges

Challenge 1 – Stock diversity management 

We need to face some challenges when creating product sets. The first challenge is stock diversity in the product feed. 

Products often have different sets of attributes, which makes it difficult to consider them all when creating filters at the same time in only one AI campaign. 

stock diversity in the product feed example

For Leroy Merlin we added to the product feed additional attributes dedicated to product sets, which aggregated products from the same area. For example, products of the same model, the same color and the same series. 

Thanks to this, we can easily fit the results to those attributes and display the relevant recommendations. 

Also, we created separate AI campaigns for various product categories, which enabled us to add different filters adjusted to the various categories. Separate AI models often help us to meet the business needs regarding product set campaigns. 

Challenge 2: Including products in related categories 

In order to display a relevant product taken from the proper categories, we tend to use separate models with proper settings of the categories.  

  • including top category: This concerns products set in the highest subcategory in the hierarchy
  • ncluding parent category: This concerns products set in the higher and the highest subcategory
  • including current category: This concerns products set in the same category as the context product
  • including custom category: This allows to set any category defined by you

It all depends on your business needs in a certain category. 
We can exclude a current or top or even custom categories as well. It all depends on the stock on the category and on the variety of products. 

Proper filters are very important, sometimes one campaign will be enough based on the proper AI model. However, it's basically the matter of stock and the structure of attributes in the product feed.  

Our real Use Cases 

Use case 1: Decathlon 

Decathlon is one of the biggest sporting goods retailers in Europe. We prepared advanced product sets for them with a few interesting elements. 

example of advanced product sets with a few interesting elements

Above we see the main product, which is an MTB bike and product suggestions that complement the bike. Those three products are complimentary recommendations from categories which we are able to define.  
As you can see, you can choose the product at your own discretion and compose your own sets. You can also add products separately or as a whole set with one button.

Of course, it’s completely flexible. If you want, you can add some restrictions that will make the recommendations show only products from a particular category. 
Read more how to build personalized product set in Synerise 

Use case 2: Leroy Merlin 

Product sets are also used by Leroy Merlin, one of the retailers of building materials and interior accessories. 
The difference is that here we are using the recommendation of complementary and visually similar products to compose sets.

Example of the recommendation of complementary and visually similar products

Here we have a blue curtain, and a very accurate recommendation of accessories that will complement your room because they match the color. 
This is very flexible, and the final look of your recommendations can vary on your business needs.  
Read more how to build product set with decoration category in Synerise

Summary 

Building product sets and offering more products together gives you the possibility to increase the total value of your basket. Also, in Synerise implementation of those cases is very flexible. Try to build such recommendation yourself and we are sure that it can help you to easily achieve better results.