• Homepage
  • Blog
  • Create the Perfect Customer Journey Based on AI Recommendations

Create the Perfect Customer Journey Based on AI Recommendations

6 min read
Create the Perfect Customer Journey Based on AI Recommendations

Personalization is the key to success e-commerce. It can be deployed in any number of ways but one of the best examples is to apply it to websites to power personalized product recommendations.

According to PwC data, consumers are likely to pay up to 16% more for a better customer experience. One way to deliver that experience is to help clients make purchasing decisions by implementing recommendations. 

You can choose from many different types, but it is important to know where each type works best and how to use them well. To boost conversions, you have to know the key rules for implementing them at all stages of customer paths on your site. 

Newbie on the home page  

The home page is the first place where the customers go. This is the first point of contact with clients and that is why at this first stage you must properly encourage them to stay here longer. 

The type of recommendation that works best here is personalized recommendations. They show products which are best suited for each customer. The model is trained on the basis of the behavior of all customers, so each customer will get a unique set of recommended products. Data used to build this algorithm includes product feed, page views, transactions and any other data about customers you are able to add to the system. 

This type of recomendations aims to show the user that your store contains products that may interest him. This makes it easier to encourage client to click on them and stay on the page longer. 

But how can you create personalized recommendations for users who visit a site for the first time? Advanced algorithms can present products that are liked by the largest number of visitors, or instead show promoted products, or preferably chosen from a given category. 

We often want to think of each of our clients as individuals, but in reality, we are all more similar than we might think. It is the same with product preferences. 
 
Therefore, an effective solution is to present the most-often chosen, personalized products displayed to most customers. Of course, the more our new user spends more time on the site or when he returns to us again, then these recommendations will become more and more accurate because they will be based on a larger amount of user data. 

Interested customers on the category page 

A customer interested in buying in our store begins by browsing our site and product categories. The list of products in a given category is also a very good place to add a box with recommended products. 
In this case, we can suggest: 
 
for anonymous users - promoted or most frequently bought/viewed products from a given category; personalized recommendations will also work. We can gather the history of a client that is not logged in. For anonymous, first time users, the personalized recommendations have a ‘cold start’ solution, which chooses products from among popular items. The text suggests that this will have to be done in two separate campaigns/recommendation types. 
 
for recognized users - personalized products best suited to the user from the currently viewed category. 
In this way, we can select several products in the recommendation bar that can help the user choose and search within the current category. 

Activated customer on the product page 

The product page is a very good place to apply many different types of recommendations. The two most relevant examples are: similar recommendations and visual recommendations
Similar recommendations show related offers and recommend similar products to the ones the customer is looking for. 
 

  • Similar recommendations based solely on the product feed for the 'cold start' of the product, when it is first added. After gaining interactions from users the model bases the recommendation also on the co-viewed products. This way the model is built not only on the product feed but also on human interaction. 

    At this point attributes can be used to filter recommendations to narrow down the type of returned products. The filters are based on business knowledge that the models simply don't have. 
     
  • Visual recommendations show the results of visually similar products because the model analyzes images of the products from the product feed. It is an effective solution especially for the clothing industry, where products look is a key criterion.Thanks to this type of recommendation, we gain a greater opportunity to persuade the customer to buy. 

    If the product being viewed does not appeal to the user for some reason, the recommendation bar may persuade him to look at another, similar one, without having to return to the list of categories and search again.

Decided customer adding product to the cart 

A customer who wants to add a given product to a basket can be described as a deciced customer, and we can treat him as a success at this stage of the conversion. 
 
This does not mean, however, that we cannot influence the size of the basket, even at this stage. We can choose two recommendation strategies that will allow us to increase profits from sales. 

For this purpose, we can use complementary products, i.e. products that belong to related categories. This means showing items that are frequently bought together or somehow complement the item placed in the cart. 
 
Knowing which categories are often purchased together, we can create more effective and effective recommendations. 

Another solution is to create product sets, which are sets of several products that can be bought at a bargain price. This gives the customer a clear benefit, because for a larger number of products (of the same type or belonging to a specific set), they can pay less, and for us it is a profit in the form of a larger basket of a single customer. 

Product sets can be determined individually to suit the nature of the item, e.g. for each camera it will be an additional lens, and for the computer headphones and mouse. They can also be the same products, but in a larger number (then the discount will automatically be e.g. 10%). We can also determine that they will be products from the same category (e.g. 3 books from a given author). 

Last call for purchase: in the basket 

Last but not least, you can add your recommendations on the checkout page.  This is where we can encourage users to expand their shopping list. 
 
The best type of recommendation that we can apply here is the cart recommendation, i.e. the products most often bought together. Based on a previous analysis of the most frequently bought together products, we can create a personalized set of products that will interest the user. 
 
In some industries, such as clothing, it is worth taking into account the purchase history of a given user in these recommendations and excluding already purchased products from the recommendation.  
In the case of FMCG, this is not necessary. On the contrary, it is worth offering in such recommendations the products most frequently bought by a given user. By reminding him of his favorite products, we can encourage him to add them to the basket at the last minute. 

Summing Up

Encouraging undecided users to buy and increasing the value of the basket itself are always primary goals. The most important thing is that the presented products are selected in the right way and contain such recommendations that may actually interest customers. 
 
That is why it is so important to collect data on all customer activities, including viewed products, so that the recommendations presented are best suited to the expectations of customers. 
 
Want to learn more? Check what else AI recommendations can do on our Use Case section!