• Homepage
  • Blog
  • AI or Industry Knowledge – Who Makes Better Decisions?

AI or Industry Knowledge – Who Makes Better Decisions?

8 min read
AI or Industry Knowledge cover photo

Recommendation systems help users to take correct decisions in their online transactions while increasing sales and engagement with the brand. On the other side we have industry knowledge and experience, which is also very important. How do recommendations succeed in this competition? How can you use them to really boost your profits? In this article we be your guide through the word of recommendations and show the new possibilities that this world can offer.

Introduction to recommendations

First, let's talk about recommendations themselves. The main goal of product recommendations is to present the user's products or services based on their preferences and current needs.  

Thanks to recommendations, you can increase the average cart value in many ways.  

For example, you can offer your customer complementary products or items that are frequently bought together.  

You can also offer more expensive products in the category that interests customers, and you can finally offer them a similar product, hoping that they will be tempted to buy. You can also improve your conversions with recommendations. When customers find products that are best suited to their needs, they are more likely to buy.  Thanks to real time data processing, our recommendations are a very flexible tool and adapt to the needs of the user and change over time with them.

You can build long term relationships with your clients because our system can recognize and remember behaviors, page visits or transactions. Thanks to this, we can use that information in our recommendations to serve our clients the products that are best suited to their needs.

Types of recommendations

In Synerise we have a few different types of recommendations:

  • Similar products
    This is the easiest type of recommendation to implement because its only information source is from the product feeds. The main purpose of this recommendation is to help customers make purchase decisions faster by showing related offers and recommending similar products to the one that they have.
  • Visual similarity
    In this type of recommendation, we show products that are visually similar (shape, color, style, etc.).
  • Cross-sell recommendation
    The purpose of this type of recommendation is to offer additional products to the ones the customers are viewing and get them to add more items to their shopping carts. The system compares currently viewed products with products that were viewed by other customers with the same interests.
  • Cart recommendations
    This type of recommendation allows you to create product offers based on the products your customers have added to their carts. The algorithm analyzes which products have been added to the cart in similar situations by other users.
  • Personalized recommendations
    Finally, we have the most interesting type of recommendation; personalized recommendations. This type of recommendation lets you suggests the products based on user buying preferences and their behavioral profile. In other words, users always see content and products selected on the basis of what they have previously viewed or bought and an analysis of the product feed.

Requirements

The most important element to start with is implementing the Synerise tracker to be able to gather data about page visits and customer transactions.

You also need to properly implement products to your website using product feeds and use OG:tags.

Read more about Product Feed >>

Read more about OG:tags >>

There are also a few things that are optional but can make a huge difference. You can gather historical data about page visits, transactions and other activities. They should be from a time period of at least three months, but of course more is always better.

Remember that every product must have a unique product ID – the same applies to the product feed, OG:tags and on the website.  

Of course, you won’t need all of those things for every type of recommendation at first. But if you want to start using personalized recommendations you have to think about other data that can be helpful.

How to create recommendations

At first you have to choose what type of recommendation would you like to use on your website. All of them were described above.

The next thing is to declare the quantity of returned products in your recommendation.

The next important step is to configure the recommendations filter. You can use a brand filter, category, price, discounts, gender or something else.

The preview button is important here – you can use it to preview the recommendation that will be displayed to a specific customer (personalized recommendations) or for a specific product (e.g. similar products, visual similarity). As you can see, this is a very useful functionality that will allow you to fit your recommendation to your needs. You can also test every model recommendation in this way.

The last step is to adjust the additional settings. This part will help you with choosing the optimal set of products and put them in the best order.

When you build your recommendations, you should ask yourself how you want to display them to your users. You can add recommendations to emails, web pushes, mobile pushes and on the website as dynamic content. Decide what kind of campaign you want to add the recommendation to.  

Which type of recommendation works best?

You can probably determine exactly which products sell best in your own store.

However, a lot of clients are often curious about whether our algorithm can really be smarter than their specialized industry knowledge. We have asked ourselves this question many times and we have tested our algorithm of personalized recommendation in many different combinations. So who will win this fight?  

Let's check it together and find out the real results based on the story of one of our clients from the fashion industry.

Our customer initially tested various types of recommendations to find out how well they worked. Based on this in-depth analysis and testing, we decided to immediately use the model of personalized recommendation on their website. However, the client wanted to use cross-selling recommendations, offering his clients some other selected products based on their experience.  

We agreed to this challenge and started to test personalized recommendations on the basket page.

 

Version 1: For suit shirts – cross-sell recommendations of ties

Version 2: For suit pants – cross-sell recommendations of suit shirts

Version 3: For other products - personalized recommendations

The results:  

Ties for Suits

  • CTR: 2%
  • Conversion: 1%
  • Average cart value: 45.30

2% of users visited the product page directly using the link from the recommendations sets, and only 1% of customers who saw our recommendations made a purchase. So as you can see the results from the first manually driven set were not very impressive.  

 

Suit shirts for suit trousers

  • CTR 3%
  • Conversion: 0,6%
  • Average cart value: 107.20

If you compare the results for the second set, you may notice that our CTR here has increased. You might also notice that our average cart value is much higher than before. We have over hundred here instead of 45 before but unfortunately the conversion rate is lower.  

So to sum up, if you want to compare those two results prepared manually according to the best knowledge of the client, the results were not very satisfying.  

Personalized recommendations

  • CTR: 6%
  • Conversion: 4%
  • Average cart value: 200.10

First, we have a CTR that is three times higher compared to the first set and double that of the second. Also, our average cart value is impressive here. This conversion rate is definitely much more satisfying for us at 4%.

Check our use case to discover how to do it >>

Summary

We can confidently say that e-commerce owners know their business very well but we are also convinced that algorithms will be always one step ahead of them. It is definitely worth trying to use them to boost the profits by present more adequate recommendations to customers.