What is an e-commerce recommendation system? Everything on the topic!

You ever heard of recommendation system? Today, it is an indispensable technology for any e-commerce, as it makes product recommendations for the consumer, acting as a virtual seller. This facilitates navigation within the store and raises the user experience, because it makes the shopping journey much more enjoyable. 

But recommendation systems are present in our lives in many other channels and segments. So, before anything, how about understanding what it is in general? In this post you will get a complete overview of this and EVERYTHING that involves a recommendation system for e-commerce, in the following topics:

Come on? 

What is a recommendation system?

A recommendation system, or recommendation mechanism, is a tool that uses a series of algorithms, data analysis and even artificial intelligence (IA) to make recommendations online. These recommendations can be Custom for each user or not, depending on the objective of each platform, the amount of data obtained and even the type of technology used. 

When the focus is on personalizing the user experience, the system uses data related to that user's profile and navigation - such as clicks, ratings, and searches - to recommend items that are most relevant to him.

Above all, a recommendation system collects information and, with that, facilitates the process of decision-making when showing and recommending a selection of items. The item can be a product, a form of content or even a person - in the case of social networking sites or in the suggestion of friends on a social network.

This information collected concerns three elements:

  • the items to be recommended; 
  • the user who will be impacted by these recommendations;
  • and other users who have already interacted with the platform. 

User data can be of two types: explicit or implicit. 

Explicit: consist of information provided by users, usually in the face of any question or request, such as comments and evaluations. 

Implicit: they are produced spontaneously by users and have to do with their behavior during navigation, such as clicks, searches, time spent on a page, etc. 

The quality of the recommendations is directly related to the quantity and quality of the data obtained. So, the more data there is about a person and other people similar to him, the better and more personalized the recommendations can be, arousing that user's interest and helping in decision making. 

When did it come about and why?

The term recommendation system may seem new, but it appeared in the years 1990. His first studies originated in several areas, such as cognitive sciences, approach theory, information retrieval, forecasting theory, administration and marketing, and arose from the difficulty of people in finding and choosing items in the face of immense amount of information available on the internet.

Thus, in addition to solving a problem, the recommendation system emerged as a business opportunity, taking advantage of this large amount of data and information to generate profit in an attractive way. 

Tapestry, developed at that time, is considered by many to be the first recommendation system. He gave rise to the term “collaborative filtering”, since the filtering of information carried out in this system was done through collaboration between people. 

At first, “collaborative filtering” ended up being the name adopted to address any recommendation mechanism, but then the creators themselves began to call it more generally “recommendation system”, since not all systems need to use the technique collaborative. 

Tapestry also motivated the creation of the study area focused only on recommendation systems. From it, some researchers started to dedicate their studies to identify and solve recommendation problems related to evaluation structures. 

Where is it present?

Recommendation systems are present today in most services streamingin the social networks, ecommerce and even in app stores that we use on a daily basis.

Despite being quite distinct segments, they have in common the need to deal with a large amount of information. Then, the system helps to organize them in a more attractive way for the user, improving the EXPERIENCE. That way, even if he is not sure what he is looking for, he has easy access to several suggestions and does not need to keep looking in categories and clicking on item by item. 

The way of recommending changes according to the business and the items to be recommended. But in general, everyone works with three types of suggestions: 

  • those that take into account the user's profile and are more relevant to him; 
  • those that take into account the popularity of the item in relation to all users; 
  • those that show the news. 

See some examples below.


recommendation system Netflix it is undoubtedly one of the best known and used by people. Its goal is to help users choose movies and series from so many options. Without having to search for hours until you find something that piques your interest.

Netflix recommendation


O Spotify is one of the most used audio streams in the world and uses the recommendation system to suggest podcasts, songs and even assemble entire playlists for its users. 

Spotify recommendation system


O Facebook dominates the social networking market and uses a recommendation system to suggest new friends, recommend sponsored ads and relevant content for each user's profile.

Facebook Ads RecommendationFacebook content recommendation

Google Play Store

Google uses recommendation mechanisms on many of its fronts, but here we highlight the Google Play Store, an app store that today also works with movies and books and uses recommendations to suggest items and optimize navigation.

Play Store games recommendationPlay Store e-books recommendation

The Amazon

A The Amazon is a pioneer and one of the most notable and old cases of success when it comes to the recommendation system. 

They have used this technology since 1999, always cherishing the customization of shopping experience. Over the years, they have been improving the system more and more and today it is the most valuable company in the world: it is worth almost half a billion dollars. 

Its system is geared towards product recommendation, which facilitates the purchase journey of its consumers and helps in decision-making e sales conversion.

Here are some examples of the recommendations responsible for this successful Amazon case:

Amazon recommendation system

Amazon recommendation showcase

Most of these services have their own recommendation systems that use artificial intelligence (IA) so that the recommendations are more personalized and, therefore, more assertive. There are systems that do not use AI and, therefore, are unable to offer such a personalized experience. But we'll talk more about that in the next few topics.

Recommendation system for e-commerce

In e-commerce recommendation systems, the main objective is the suggestion of products for consumers in calls smart shop windows or recommendation windows. They can appear on the website's home page, on product pages and in the shopping cart. Shall we better understand how they work?

Smart recommendation showcases

As smart shop windows act like one virtual seller and intelligently recommend products within the online store. 

They can even recommend items in an even more assertive and personalized way than an average salesperson, due to the large amount of data and information that the system can obtain about users. It all depends on how much the consumer browsed the store and how much information he made available to the system. 

However, if he has little interaction with the site and there is not enough data, the only way is to work with non-personalized recommendations, which can also be very effective. 

See some examples of smart showcases below.


When the consumer enters the online store and starts browsing, clicking and searching for products of interest, the system identifies his profile and can make custom recommendations and more relevant. This is the case of the Diesel below, which shows products similar to what he's already looking at and quite similar to each other.

custom diesel showcase

They make more enjoyable browsing, because the consumer can see different items of interest without having to search in categories or filter information.

Not customized

If the customer has just arrived on the site and has not yet surfed enough, it is difficult to identify their interests. In this case, it can be impacted by non-personalized display cases. They are more generic and, in general, show the same products to all customers. 

The strategy here is to bet on mental triggers, such as scarcity and social approval, showing the latest offers or the most viewed and / or sold items. The launches are also widely used.

Diesel recommendation showcase

It is common for both strategies to be used in e-commerce recommendation systems, working together for all types of situations.

Learn more about smart shop windows.

Stand-alone showcases 

Smart showcases can also be autonomous, that is, they can work automatically and without the need for manual settings, being much more efficient. This happens when the system uses artificial intelligence. We will cover in more detail in the topic: Recommendation system with artificial intelligence.

Have your own or outsourced system?

As you saw in the previous topic, Amazon is a great success story, especially when it comes to the recommendation system in e-commerce. They work with recommendation windows in a own system it is costs a lot of money e it takes a lot of work. After all, it is necessary to have a technology team totally focused and specialized in this.

But, fortunately, today any virtual store, even if it doesn't have an internal team to develop its own system, can also count on a recommendation mechanism on its website, as there are third party services totally focused on that. 

Some do a specific project for each store, which takes more time and costs more. Others have a standard structure that can easily be adapted to any store. 

If your company is not Amazon, the best way is to opt for an outsourced system. At the end of this text, we list some points that will help you when it comes to choose a recommendation system to your store. Check out!

Types of recommendation systems

There are several types of recommendation systems, all of which work with data, but in different ways. Some use only statistics, information retrieval (IR) and data filtering. Others use these methods in conjunction with artificial intelligence, in machine learning techniques. 

A process that is on the rise today and is widely used is feature extraction, a segment that uses artificial intelligence to extract features. Posteriorly, the extracted data is processed by filtering or used in other IAs. And there are also systems that use intelligence together with filtering to make recommendations. 

In general, all recommender systems go through 4 main steps. Check it out below.

1) Collection: collecting data is the first necessary step for a recommendation mechanism to work, as they are the foundation of everything. As already mentioned, these data refer to the items that are intended to be recommended and to the users that interact with the system, which may be implicit or explicit. 

2) Storage: then, it is necessary to store the data in some way so as not to “lose it”. You can choose a NoSQL database, a standard SQL database or another type of storage. The important thing is to store them in an organized and structured way and according to the need of the system and the amount of data. 

3) Processing: another very important stage is the processing of data, that is, the stage in which they are analyzed and ordered according to their characteristics, facilitating the next stage of filtering.

4) Filtration: it is at this stage that the recommendation actually happens. In this phase, the data collected, stored and processed are filtered in order to generate the most appropriate recommendation for each situation. There are several ways to filter data, it all depends on the type of algorithm used.

Content-based filtering

In this type of filtering, the focus is only on user concerned and the characteristics and similarities between content or items.

This happens when a product is recommended based on the last search or purchase made. If the consumer bought a science fiction book, for example, the system assumes that another science fiction book, with similar characteristics to the previous one, also has the potential to be purchased by him.

Some of the problems with this approach are the content bubbles and the need for large amounts of data.

Content bubbles: the system always recommends the same types of items. This lack of diversity, in addition to the risk of making navigation boring for the user, does not increase his consumption pattern. 

Large amount of data: another challenge in this type of approach is the need for a large amount of data, both from the user and from the items. 

Collaborative filtering

In the collaborative approach the focus is on the relationship between all users and os items. In this case, the similarity between one item and another is defined based on the opinion of users. It is as if they share information about the products and collaborate with each other, hence the name. 

The essence of collaborative filtering is as follows: two users who have liked the same item at some point will potentially like another similar item in the future. 

This filtering is used on occasions when little information is available about a specific buyer. Therefore, the algorithms collect data, such as evaluations and previous purchases, and combine this information with other users who have similar preferences. Thus, they assume which products are most likely to be relevant to him. 

For example, if a person enters an online store and searches for "shoes", but eventually ends up buying a bag, the system identifies a correlation between both actions and categories. Then, as more people repeat the same actions, a product recommendation is created for profiles with similar behaviors. 

Hybrid filtration

Hybrid filtering combines content and collaborative approaches. It is very effective, because it brings together the positive points of each filter. 

Generally, the rankings for each approach are analyzed separately and combined later. In other techniques, the two approaches are combined in the same framework. 

Other approaches

In other approaches, nothing or little personalized recommendations are used. This is the case demographic filtering, whose selection criteria is based on demographic attributes, such as age, gender and location. In other cases, non-personalized algorithms recommend items in “releases”, “most viewed”, “promotion”, which are generally the same for any user. 

The goal is to solve the cold start. That is, when a new user or item appears in the system, with few interactions, and it becomes more difficult to arrive at personalized recommendations. So bet on these more generalized approaches it is a way to suggest something even without having so much information.

It is a way to encourage the user to start browsing and then inform the system about its characteristics and profile. It is also a solution for items that have little interaction and need to be exposed in some way.

Recommendation systems with artificial intelligence

All the types of approach mentioned above can be worked on together with the artificial intelligence. The two major differentials of a recommendation system with artificial intelligence are the customization and automation


As much as systems without AI seek to personalize the recommendations, they will not be as effective, as they only meet fixed and linear rules. 

Already artificial intelligence one of its main characteristics is non-linearity, as it uses non-linear calculations in order to work as close as possible to a human brain, through intelligent algorithms. Therefore, can further customize the experience.

Systems without artificial intelligence, in turn, use simple algorithms. With the simple mathematical algorithm, recommendation expectations are more predictable and limited, as they meet a specific rule, which will always work the same for any user and item.

With intelligence, even more generic storefronts like "launches" and "best sellers" can be customized.


Um recommendation system with artificial intelligence it has the advantage of being automated, since AI does real-time analysis and not just A / B testing, as it happens in other systems. This makes processes much easier e optimizes time team's. 

An example in electronic commerce is the freestanding showcases, which work automatically. Based on real-time analytics, artificial intelligence identifies which storefronts are converting more, which is the best position and which are the most relevant for each user. 

Thus, they position themselves automatically, without the need for human interference. isso decreases the need for labor to perform any type of configuration. 

To learn more, read this post that talks about it: Standalone virtual showcase: how to customize automatically.

It is because of personalization and automation that the biggest cases of recommendation systems, such as Amazon and Netflix, use artificial intelligence. 

Recommendation templates

In electronic commerce, there are many types of recommendation showcases, for the most diverse strategies. Below, we selected some examples of recommendation templates most used.

Best sellers

best selling showcase


showcase launches

Personalized recommendations

recommended showcase for you

Frequently bought together 

recommendation showcase often bought together

Similar Products

showcase similar products

To learn about other recommendation models and to learn more details about each one, read also: 8 product recommendation templates for your online store.

Which recommendation templates to use and how?

To find out which recommendation templates to use it is important to first get to know your audience well, through research and analysis of results. It is also important analyze the performance of each showcase after active, to find out which ones work best with your audience and where. 

In recommendation systems in general, A / B tests and, from there, analysts identify and configure showcases in places where there is a greater chance of converting. 

In those using artificial intelligence, freestanding showcases they do it automatically, since, as we mentioned earlier, intelligence performs real-time analytics.

But in addition, it is also necessary to use each one in the right place and for the most appropriate strategy, check out some tips.

At the store home

The most widespread storefronts, such as "Launches", "Best sellers" and "Most viewed" generally work well on the website's home page. As well as personalized showcases, as long as the customer has already browsed the store enough and revealed their interests.

On the product and shopping cart pages

Showcases that suggest a joint purchase or complementary products make more sense on the products page and in the shopping cart. 

While the showcase of similar products works well on the product pages, but not on the shopping cart, it can leave the consumer confused and undecided between the products, instead of making the purchase. 

Showcases with the face of your brand

Another possible thing to do is to be creative in the titles of the showcases, using a language closer to your customers and more like your brand, like the example below, from stoned shop

stoned shop window

Benefits of having a referral system

The main purpose of a recommendation system is to raise the user experience during navigation and, consequently, generate good results for the business. Therefore, in order for you to understand the benefits of this technology for e-commerce, we list the benefits for the final consumer first. Check it out below.

Benefits for the final consumer

1) Facilitates the purchase journey

The recommendation showcases optimize the search for the product within the virtual store, as the customer can view various products while browsing, without having to search for categories and filters. This ends up facilitating the purchase day.

2) Provides more assertive purchases

When the system suggests products related to the consumer's taste and profile, the greater the chances of a more assertive and successful purchase. 

3) Promotes productive purchases

The suggestion of complementary products allows the customer to buy something that he had not even thought about purchasing, but that will work well with the desired product initially.

4) Customize the experience

The personalized windows make the customer feel that he is in a store specially organized for him. It makes you feel important and well taken care of. 

5) Enables the discovery of new products

When viewing different types of recommendations during the purchase, the customer has the possibility to discover new products in a simple and relaxed way.

All of these benefits contribute to an excellent shopping experience, making the customer much more satisfied.

Benefits for the online store

When the customer is satisfied, the benefits for the store begin to appear.

1) Increased conversion rate

This is because the easier it is for consumers to find what they are looking for, the greater the chances of conversion and effecting the purchase.

2) Increase in pages accessed per session

The recommendations make the user click on several products and, with that, access more pages while browsing.

3) Better placement in search engines

By accessing more pages per session, the customer ends up spending more time on your website. And the more that happens, the better your site will be ranked by search engines.

4) More traffic to the site

If your site is better positioned in search engines, naturally you will receive more traffic. Furthermore, by having a good shopping experience, the chances of your customer coming back and recommending it to other people are great, which also ends up increasing traffic.

5) Increase in average ticket

With the incentive to buy in pairs and suggest items with higher purchase value, it is possible to increase the value of average ticket your store.

6) Competitiveness

Having a recommendation system is almost a watchword for online stores, since most have this technology. So, to gain competitiveness in the market, it is necessary to have a good recommendation system and offer the best possible experience for the customer.

7) Greater customer engagement

By having a good shopping experience inside the store, the customer becomes more engaged and, with this, wants to promote your brand and tell everyone how well it has been served.

8) Loyalty

When the customer has a good experience and is more engaged with your store, loyalty is a natural consequence, as he trusts your brand and will want to repeat the dose whenever possible. 

Recommendation systems in conjunction with other tools

For even better performance, a recommendation system can work in conjunction with other tools, further elevating the shopping experience and guaranteeing even better results for your e-commerce. Meet some of them.

Smart search

A smart search is a tool that uses artificial intelligence to optimize searches online. In e-commerce it is widely used in conjunction with the recommendation system. 

Although the shop windows facilitate navigation and the shopping journey, there are customers who prefer to go directly to the search field to search for what they are looking for. These are the most determined customers, so it’s important to take the opportunity and show the right products quickly and efficiently. And that is what smart search does.

Generally, an intelligent search includes:

As artificial intelligence is a technology that is always evolving and developing, the more time passes, the more capabilities the intelligent search can have. 

SmartHint is an example of a recommendation system that works in conjunction with smart search. To learn more about this technology, read the post: What is smart search and why have it in my e-commerce?

Retention Pop-ups 

Os retention pop-ups they are another tool capable of further enhancing recommendation systems. Its main strategy is avoid abandoning the shopping cart during the heat of emotion, instead of waiting for the abandonment to happen and then trying to reverse the situation.

SmartHint also works with this solution and has pop-ups that bet on urgent mental trigger: show a flash sale with countdown timer for the product being viewed. This draws the customer's attention and makes them want to close the purchase so as not to miss the offer. 

Want to better understand how retention pop-ups work? Then access this content: Retention pop-ups for e-commerce: what they are and how they work.

Cart recovery 

Cart recovery through email marketing is an option of a complementary tool widely used in conjunction with the recommendation system, in case it is not possible to avoid abandoning the cart. 

O You sent is a great and very complete alternative, as it monitors in real time and sends emails automatically as soon as the person leaves the checkout page. It also brings together other interesting solutions, such as billing collection and cashback.

Billet collection

Having a billet collection tool is important, as many customers, even after they have closed the purchase, end up not paying their bills for sheer forgetfulness, for not having all the necessary data in hand or simply for having given up. 

So, even if the recommendation system has worked and facilitated your customer's purchase journey, these little details can put everything to waste. So, if you can, count on such a tool as well.

Sales and after sales automation

The recommendation system guarantees a good experience during the purchase, but remember that the shopping experience it also involves everything that comes before and after. You must be consistent and provide your customer with impeccable service at all stages. 

For that, a good request is to have sales and after-sales automation tools that give more alternatives to customers, streamline processes and optimize the work of your team. 

A Omnichat is an omnichannel sales platform, with which the customer can make payment directly via WhatsApp or via chat application. And it is also possible to integrate catalogs, prices, inventory and products with e-commerce.

Already Aftersales works with the automation of the entire after-sales process, with it the exchange operation is much simpler and faster, as the customer can access, within his e-commerce, an interface available to make the exchange alone, without e emails and long calls.


A cashback tool is interesting because it offers one more opportunity for the customer, which goes beyond the purchase and ends up exceeding his expectations. The conventional cashback returns to the customer part of the investment made in the purchase. Is way to encourage you to consume

Enviou takes advantage of this opportunity and offers cashback at strategic times, such as cart recovery and billet collection. 

Already Poland Work as social cashback: allows the client to choose between two social causes for his company to make a donation. As it appears at the time of finalizing the purchase, it is also a great way to prevent cart abandonment, as it delights the customer through empathy and social commitment

What to analyze when choosing a recommendation system?

Having a recommendation system within the virtual store is undoubtedly a necessity. But among so many options, how to make the choice? We have separated some important points of attention for you to observe when making this decision.

Integration with your platform

First of all, check if the recommendation system analyzed has integration with your platform. If not, check the possibility of carrying out the integration, what is the necessary procedure and if it is worth waiting for. 

Implementation time

Can be that even with integration with your platform, the system takes a long time to be implemented in your virtual store. There are players that take months, because they use old technologies that need a specific project for each store. On the other hand, other systems use more advanced technologies that can scale, as they have a ready structure. Thus, it can take a few minutes to implement. This is a very important point to be analyzed, because time is money. So the sooner your system is up and running, the faster your store will experience results. 

Cost benefit

For the same reason mentioned in the previous item, some recommendation systems are very expensive because of the low technology, which requires more time and more work to be implemented. More modern technologies, such as those that use artificial intelligence and cloud computing, can be much more accessible in terms of cost and still have a much higher performance. 

Success cases and results

Check the success cases of the systems you are analyzing, who are the customers using the tool and what were their results. Tip: enter two stores (or more) that use different systems and browse their websites. Simulate an online purchase, click on products, search and analyze how shop windows behave. That way you will be able to see in practice more or less how the recommendation system will work in your store. 

Results analysis

Also make sure that the system to be hired has a results analysis mechanism, with which you can monitor the performance and evolution of the tool within your virtual store daily and in a simple and clear way. Only then will you know if the investment is paying off and / or if you need to bet on some other strategy.

Complementary tools

Observe if the system you are analyzing works with complementary tools, such as those mentioned in one of the previous topics. One charges more and how much for that. Or if you have any partnerships with other players that allow you to use them. But, in addition, check if these tools really work, because there is no point in having access to tools that perform poorly. 


Another important factor to be analyzed is how much the contracted recommendation system will generate the need for maintenance and manual configurations. Some systems are 100% automated, this generates less work and less labor costs.

We hope that this content helps you and serves as a guide in choosing this technology for your online store. Good sales!

Discover SmartHint's recommendation system!

Written by: Tania d'Arc


Rodrigo Schiavini

Rodrigo Schiavini

Founder and business director of SmartHint, he is also regional director Paraná of ​​ABComm, with more than 10 years of experience in electronic commerce for major brands in the most varied segments.
latest posts

Contents and strategies
for e-commerce

Receive specialized content in your email.