top of page

Building online 
recommendation engines to improve 
customer experiences

E-commerce product inventories and availability have grown rapidly. However, if customers can't find what they're looking for on your website, they'll leave the store and buy elsewhere. Better search capabilities on your site are one option, but unfortunately, customers don't have the time to dig through search results or create elaborate filters. But there's another option.

The Challenge

Related product recommendations provide the customer with options for related products that are relevant to his interests and purchase intentions. He could be interested in an item that is out of stock in the size he wants, but for which there is a similar item in the right size. Product recommendations can make that connection for the customer and combat lost sales opportunities for your business. The right product recommendations can also increase cart size and average order value by connecting complementary products.


However, many companies don't have the time to sift through their entire inventory to make these recommendations.

The opportunity for deep learning

Deep learning models empower machines with the ability to understand the content of images and text. These models allow us to automate repetitive tasks, such as matching an entire product inventory, at incredible speed and scale. But understanding what an image is, such as a blue shirt, is only half the battle when it comes to making appropriate recommendations to customers. How can we train a model to find similar products based on that understanding?

Suggested solution

Customer product recommendations can be based on a wide range of important information. Having an understanding of who and what your customers like will always be the foundation for building successful AI-powered product recommendations. Should recommendations be based on similar product descriptions, product images, similar brands/qualities, or perhaps similar customer profiles?


In this case, we are going to focus on the image and text descriptions where Deep Learning has its strengths. Using text and image embeddings, a Deep Learning model can identify the similarities between product images or product descriptions. It does this by learning to create representations of these texts or images and then mapping them to similar representations that the model has learned from.

Data requirements

Your best starting point for the data to create a similarity model is your product inventory. Whether it's a simple solution like an Excel file or a more advanced system like a PIM (Product Information Management), you should be able to create a data set with the product images and/or descriptions you already have. It is important that the images are the same size and the descriptions are detailed so that the model gets a good understanding of them. Remember, the better the data, the better the recommendations. So it's worth investing time and effort into creating a well-formatted data set.

How does the model work?

Models of similarity operate in the world of representation. This means that the model searches the given data set and assigns, for example, a mathematical representation, a vector, to a given product description. Based on this representation, it can then calculate the mathematical distance between the other assigned representations to determine how close or similar they are. A successful result would not imitate someone going through a list or exact matches, but would show a deeper understanding of the text. A good example would be:

          This coat is the best winter coat to buy this season.

          Purchasing this coat will make sure that you stay warm                   in cold 

You need some help to discuss your business case?

Contact our experienced
Sales Team now


bottom of page