Personalization of the online store with machine learning
A unique shopping experience is a value that every online store manager should bear in mind in order to keep customers and constantly increase conversions. Various forms of personalization are used in many industries for numerous purposes: starting from marketing through streaming services to eCommerce.
The main role of personalization in an online store in the year 2021 is to boost sales and improve the user experience (UX). It also works as a great way to engage undecided customers, reduce the number of abandoned shopping carts or increase the number of transactions among first-time store visitors. In other words - a modern store is a personalized store.
Personalization vs Customization
At the very beginning, we need to distinguish between two terms: personalization and customization. In Polish, both terms are translated as the personalization, although the basic principle behind these solutions is completely different.
Customization - the user is able to individually choose the options, which have been designed to provide a more personal shopping experience. For example, users who visit the store's website for the first time may be asked what category of products they are interested in so that the offer may be adopted to their individual choices.
Personalization - The system decides what content or offers are displayed on the page. The selection of the message is made automatically, based on the history of the user's behavior. This is possible thanks to artificial intelligence (AI) and machine learning (ML) technology, which, together with the information collected about purchasing preferences from previous sessions, is able to select personalized content in real time.
In this article, we will focus on the latter solution, i.e. personalization AI/ML.
Real-time personalization of eCommerce sites
Increasing number of online stores pay more attention to building a good user experience (UX) on the website. A properly tailored offer is a quick way to finalize the transaction, so it is recommended to make sure that the shopping path is personalized at every stage.
eCommerce personalization is precisely the process of providing a unique experience in an online store thanks to dynamic content, product recommendations and special offers generated on the basis of behavior, purchase history and other data referred to below.
According to a study by Bloomreach, online stores see about 20% increase in sales after the introduction of personalized content. At the same time, as many as 80% of customers are more likely to make purchases in an online store that uses some form of personalization, and 77% of consumers would recommend a brand that provides a personalized shopping experience.
It is also an efficient way to build and maintain a long-lasting relationship with customers - according to the same study, as many as 44% of consumers declare that after shopping in a store where personalization is applied, they are more likely to become regular customers.
How AI / ML technology supports online store personalization
The introduction of personalization based on AI / ML technology may seem like a complicated process that requires the involvement of IT specialists and a high-cost investment. You might think that this is an option reserved exclusively for the largest online stores. Nothing could be more wrong! Effective and easy-to-install solutions are appearing on the market, available for every store with equally advanced online shopping personalization options.
If you are wondering how it is possible that when a 45-year-old woman visits a clothing store, she sees different content and an offer on the homepage than a 25-year-old man visiting the same store? It is the real-time personalization that is responsible for this.
It is possible thanks to the information collected by cookies. Thanks to cookies, the content of websites is tailored to the user's preferences and individual needs.
Important! Cookies used for personalization do not collect personal data. The data analyzed by the AI system does not make it possible to identify users.
In order to create personalized content, the collected data should be immediately segmented into categories such as:
- Demographic data (age, gender, education, etc.),
- User behavior on various channels (websites, mobile devices, social media, email and others),
- History of previously viewed pages in the store,
- Purchase history in the store,
- Geolocation (country, city, province),
- Time (hour, day of the week, day of the month, season of the year).
Thanks to the appropriate segmentation, the offers on the store's website can be tailored to a specific user depending even on the time of day.
Personalized content on the store's website informing about, for example, a current promotion or sale, may appear in various forms. The most popular formats are: banners, pop-ups, inline content, surveys / forms, infobars.
However, when creating content you should not forget about the specific context the target user operates in. It is, therefore, worth addressing the following questions:
- What prompted the user to visit your site? Was it an advertising campaign? Social media? The newsletter? Or maybe the search engine?
- Has the user visited our site before? If so, how often does he come back and what does he buy? Maybe he buys new monthly lenses once a month? Or does she regularly shop for groceries online during the first week of the month? Welcome the user with a tailored offer.
- Who is this user? You may use the previously segmented data on demographics, interests or geolocation, mentioned above.
- How much time does the user spend on your website? Which products and categories do they click most frequently? Recommend items that suit their preferences and tastes.
Knowing the circumstances of a potential or regular customer's appearance on your store's website, it will be much easier for you to create personalized content that the AI system will select for a specific user and display it in the right place and at the right time.
Product recommendation system
One of the most effective and easiest forms of personalization to integrate with an online store is the purchase recommendation engine. The AI / ML algorithm filters the data in such a way that a specific user will be presented with a personalized set of items that are likely to arouse his interest.
How does the system know this? The recommendation engine collects data using the cookies mentioned earlier in the article. Based on this information, it is able to generate a set of personalized purchase recommendations for the user.
Products and songs - how do the greatest do it?
The largest online stores have been using the advantages of personalized purchase recommendations for years, creating their own advanced engines. It is Amazon that stands out the most, as 35% of its revenues come from recommendations.
The success lies in the fact that its algorithm generates many different models used depending on which user will visit a given page. Recommendation models such as "others bought" use up-selling strategies and are applied in the Amazon store on the cart page. On the other hand, models such as "similar" or "recently viewed" positively influence the engagement of the store's customers on the product page. We write more about the types of recommendations below.
The recommendation engine is used not only in e-commerce. The Spotify streaming service can boast a particularly distinctive feature compared to other similar solutions. The users of this music platform are recommended artists and songs that suit their musical taste. The high accuracy of Spotify's recommendations is due to a comprehensive approach that combines three types of content filtering:
- Collaborative Filtering Model – it analyzes the behavior and preferences of a given user and compares them with others.
- Natural Language Processing Model (NLP) – it is responsible for analyzing song lyrics, artist descriptions, and genres.
- Audio model – it can analyze audio tracks in search of similarities in beat, speed, and instruments used. It is a similar technology to the one used in facial recognition tools.
Spotify's recommendation system brings impressive results and can become an inspiration for other industries, for example eCommerce. During the five years since the engine was started, listeners listened to over 2.3 billion hours of music from automatically generated Discover Weekly playlists.
What are the benefits of product recommendations?
The AI / ML technology is developing at an astonishing pace, therefore there are more and more external solutions available on the market, the integration of which does not require technical expertise in the field of Big Science. More and more stores decide to introduce this form of personalization in their store and count on the following benefits:
- Sales increase by 5-10%
- Reduction of bounce rate
- Increase in CTR
- Reducing the number of abandoned carts
- Increased trust among customers
- Easier store navigation
- Improving UX
- Recovery of conversions on unavailable product pages
- Greater customer involvement
What are the types of recommendation?
Every day we come across various types of purchase recommendations in online stores.
The most popular are based on:
- content filtering method - the system compares the descriptions of two products, then looks for similarities and generates "similar" suggestions,
- browsing history - the engine creates a list of "recently viewed" items.
However, there are many more recommendation models and their choice depends on the location on the store's website. The Recostream recommendation engine offers as many as 8 models that effectively match products to customer preferences without the problem of the so-called cold start. In other words - recommendations appear from the moment you turn on the engine on the store's website.
1. AI-Driven Maximized Conversion
It displays recommendations with the highest potential for conversion for a given product and customer. This model is automatically optimized based on actual traffic and customer behavior. It uses statistical and machine learning methods and thanks to the so-called sampling selects unique recommendations.
2. Most Viewed in Category
It shows the most viewed products belonging to the same category as the product on which we are displaying this recommendation. The algorithm generates recommendations based on information about the history of users' sessions.
3. Bestsellers in Category
It displays the most frequently added products to the cart belonging to the same category to which the product on which we display this recommendation belongs. The algorithm generates recommendations based on information about the history of users' sessions and their purchases.
4. Bestsellers in Store
It presents the products that are most frequently added to the cart among all products in the store. Recommendations are selected automatically by an algorithm that collects real-time information on which product is currently added to the cart the most in the entire store.
5. Most Similar in Category
It displays products similar to the product on which we display this recommendation. Similarity is determined by comparing descriptions, names and other characteristics of products using special algorithms.
6. Recently Visited in Store
This recommendation is only displayed to users who have previously visited the store. Products they viewed during one of their previous visits are presented.
7. Others Also Viewed in Store
It presents products that other users have viewed in one session together with the product on which we are displaying this recommendation. This model is based on the statistical analysis of the product browsing history in the store.
8. Rule-Driven Recommendations
It shows the products selected by the store owner. The person administering the online store has the ability to indicate oneself which products will be displayed at a given moment and in a given place on the store's website. This solution is particularly popular during special promotions.
Where to place recommendations?
Based on the implementations in online stores, we observe that the most favourable places for recommendations on the online store's website are the following (starting from the most profitable):
|Recommendation location||The best models for a given location|
|Product page||AI-Driven Maximized Conversion, Recently Visited in Store, Most Similar in Category|
|Category page||Most Viewed in Category, Bestsellers in Category|
|Pop-up after adding to cart||Most Similar in Category, Bestsellers in Store, Bestsellers in Category|
|Store home page||Bestsellers in Store, Recently Visited in Store|
|Cart||Rule-Driven Recommendations, Bestsellers in Store|
|Blog post||AI-Driven Maximized Conversion, Most Similar in Category (in this case, the product category the blog post is about).|
Personalization based on AI / ML technology is a solution that is increasingly common in many industries. For the e-commerce world, it is a simple way to increase customer trust, customer engagement and improve user experience (UX). The introduction of various forms of content personalization also positively affects the increase in conversion, sales and CTR.
There are numerous forms of personalization, but the most available and more and more often used by online stores is a system of personalized AI / ML recommendations. Ease of integration with any eCommerce platform without the need for technical knowledge, the ability to customize the appearance, and great freedom in choosing how to display personalized shopping proposals are the main reasons why you should try this form of personalizing your eCommerce store.
Author: Zuzanna Pajorska, Marketing Manager, Recostream