[2023] What is a recommendation engine? Mechanisms, points to note when introducing, and summary of major services

A recommendation engine is a system that displays products and information that match the user’s preferences, such as suggesting “recommended products for you” on EC sites and posting “articles you want to read together” on news sites. User convenience is improved because they can quickly access products that they are interested in and the information they are looking for. If you feel the need to deepen relationships with new users and existing customers and improve sales efficiency, we recommend introducing a recommendation engine to your EC site.

In this article, we will explain how the recommendation system works, points to note when introducing it, and recommended services.

recommendation engine

Table of Contents

What is a recommendation engine?

What is a recommendation engine?

A recommendation engine is a system that displays the most suitable products and information based on the user’s behavior history and hobbies.

For example, if you open the top page while logged in to Amazon, you will see information such as “Recommended products for you” and “People who bought this product also checked this product.”

In addition, Netflix, a video distribution service, proposes “recommended videos” based on the user’s account information and viewing history.

By introducing a recommendation engine, users can efficiently access new products and information that match their interests. Improvements in site convenience and UX (user experience) will also bring benefits to companies, such as improved sales efficiency and increased LTV.

Basic mechanism of the recommendation engine

There are five types of recommendation engine mechanisms.

  • collaborative filtering
  • Content-based recommendations
  • Rule-based recommendations
  • Personalized recommendations
  • Hybrid recommendation

It’s important to understand the differences in features, as each system presents products and information in different ways. I will explain them one by one.

1. Collaborative filtering

Collaborative filtering is a method of displaying recommended products and information based on historical data such as product purchases and web page browsing. Collaborative filtering can be further classified into “item-based filtering” and “user-based filtering”.

Item-based filtering

Calculate the similarity of each product from the user’s action history. If many of the customers who purchased product 1 also purchased product 2, the system recommends product 2 when a new visitor puts product 1 in their cart. The degree of relevance between products is the basis of the system.

User-based filtering

Suggest products based on similarity between past and new visitors . If Mr. B, whose user attributes and interests are very similar to Mr. A’s, has purchased product ① in the past, the system recommends product ① to Mr. A. Unlike item-based filtering, the system is based on user-to-user relevance.

2. Content-based recommendations

Content-based recommendation is a method of grouping multiple products in advance and making proposals.

For example, let’s say that the grouping is for product 1, product 2 with a similar design, and product 3, which is a different model of the same brand. Then, when the user accesses the detail page of product 1, pre-configured products 2 and 3 will be displayed in the recommendation feed.

Products to be grouped can be specified arbitrarily, so recommended products can be displayed even when AI has not acquired enough data.

3. Rule-based recommendations

It is a method of making rules for user actions and recommended methods for those actions in advance. Unlike content-based recommendations, rule-based recommendations can specify user behavior in addition to products.

Stores can actively promote the products they want to sell, such as proposing the most popular gifts to users who came in from advertisements related to Christmas gifts, or recommending ink cartridges from the same manufacturer to users who have purchased printers. This is useful when you want to display campaign products or limited-time products in the recommendation feed.

4. Personalized Recommendations

This is a method of suggesting recommended products based on the target user’s attributes and action history.

Collaborative filtering takes information from every user who visits your site and presents product recommendations based on their relevance. Personalized recommendations, on the other hand, analyze the attributes and behavioral history of each person and propose the best product for that person.

For example, for Mr. A, who purchased business book ① in the past, it is a mechanism to display a recommendation of newly published business book ②. Similarly, the action history of Mr. B and Mr. C who purchased business book ① will not be referenced. It is characterized by narrowing down to only “Mr. A’s preferences” and making proposals.

Personalized recommendations are suitable when you want to make polite suggestions tailored to each user, as if you were serving customers at a physical store. However, since the target user for data acquisition is only one, it takes time to accumulate the user’s behavior such as access history and purchase history as data and to be able to use it for product proposals.

5. Hybrid Recommendations

Hybrid recommendation is a method that freely combines the four mechanisms described so far. By combining multiple methods, we can compensate for the weaknesses of each method and realize a highly accurate recommendation system.

For example, if Mr. A likes to buy red products in the past, he can suggest another red product to the user as a personalized recommendation. Furthermore, when referring to other user information through collaborative filtering, we found that people who bought product ① often purchased product ②.

Therefore, at the timing when Mr. A purchases product ①, it is possible to make a hybrid recommendation such as “propose red (personalized recommendation) product ② (collaborative filtering)”.

Utilization scene of recommendation engine

Utilization scene of recommendation engine

Recommendation engines are used in EC sites, news sites, video distribution sites, etc. I will explain in detail how to use each method and examples.

EC site

On EC sites, products that are close to the user’s preferences are displayed mainly based on the user’s browsing history and purchase history.

In the case of Amazon, personalized recommendations that display “recommended for you” based on each user’s interests and hobbies, and item-based filtering based on the degree of relevance of each product, such as “people who bought this product also check” etc. are employed.

By introducing a recommendation engine to an EC site, it is possible to encourage users who closed the browser during a search to revisit the detailed page . For users, it is not only an efficient way to purchase set products such as tea cups and saucers, but it is also an opportunity to learn about new products in areas of strong interest.

News site/Job site

Recommendation engines are introduced in the form of recommending articles that meet user preferences on news sites, and recommending companies that meet the conditions on job sites, both of which are based on past browsing data. By displaying information and content that are similar to interests, it is possible to improve the rate of visits within the site or increase the length of stay .

Video distribution site

Video distribution sites use recommendation engines based on video viewing history.

A typical example is Netflix, which employs a complex recommendation system. In addition to suggesting recommended videos to people who have watched a specific video, you can see detailed consideration such as displaying the optimal thumbnail image based on each user’s attributes.

Type of recommendation engine

Recommendation engines can be classified into “ASP type” and “open source type”.

ASP TYPE OPEN SOURCE TYPE
merit

  • No need to have your own server
  • Low cost and easy to implement
  • Many types of vendors

Demerit

  • Can’t be customized
merit

  • Customizable to suit your needs

Demerit

  • you need to build your own server
  • For large companies as it requires a lot of resources and budget
  • Limited programs available

Each has advantages and disadvantages, so choose according to your budget and usage. The features of the ASP type and open source type recommendation engines are explained below.

ASP type

An ASP (Application Service Provider) is a provider that provides applications that can be used via a network, and currently most recommendation engines are of the ASP type.

There is no need for your own server, so there is no hassle, and the cost is low and the cost performance is excellent. Although detailed customization is not possible, there are many types of vendors, so you can find the best recommendation engine for your purpose.

Many recent ASP-type recommendation engines are equipped with a wealth of functions in addition to the recommendation function. Below is an example of a typical function.

  • Ranking function: Displays rankings for products with a large number of purchases and pages with a high number of accesses
  • Multi-channel support: Multi-channel personalization such as advertisements, emails, and apps
  • AB test function: Check how to display recommendations with AB test
  • Report function: You can measure specific effects from target conditions and specified rules

open source type

Based on the source code (program) that is open to the public all over the world, we build a customized recommendation system in-house. If you have someone who is familiar with programming, you can implement unique functions according to your purpose and use, and you can make detailed adjustments such as changing the layout and design .

However, it is necessary to build a unique server to operate the recommendation system, which is a big burden in terms of time, effort and cost. The ASP type is recommended if you want to easily implement a recommendation engine.

Advantages and disadvantages of recommendation engines

Advantages and disadvantages of recommendation engines

If you’re wondering whether or not to implement a recommendation engine, make sure you understand the pros and cons before implementing it. In the early stages of operation, the recommendation engine may not function well. We also tell you how to deal with that case, so please refer to it.

The advantages and disadvantages of recommendation engines are:

MERIT DEMERIT
  • Improve purchase rate, average customer spend, and CVR
  • User trust increases
  • User stay longer
  • Increase in repeat users
  • It takes time to accumulate data
  • Requires a large number of users to form a population

In order to increase the sales efficiency of EC sites, it is more effective to recommend products according to the user’s preferences and interests, rather than blindly suggesting products to a large number of targets. By introducing a recommendation engine, you can expect improvements in numerical values ​​such as purchase rate, average customer spend, and CVR.

As a result of the effectiveness of the recommendation function, users can experience the feeling of being served by a store clerk who understands their preferences well. If you build a relationship of trust with your users, they will become repeat customers.

However, it takes some time to accumulate user data. During that time, the quality of the recommendation function is low, and there is a possibility that the product or information that the user wants cannot be proposed (cold start).

In order to solve the problem of cold start, it is better to use content-based recommendations, rule-based recommendations, or a combination of both, which do not rely heavily on user attributes and behavioral history data, only in the initial stage. Once you have enough data, you can take full advantage of the recommendation engine’s capabilities.

How to implement a recommendation engine

If you implement a recommendation engine, what kind of flow should you build the system? Taking the ASP type, which is adopted by many recommendation engines, as an example, the flow until the system is introduced is as follows.

  1. Receive hearings from ASP vendors
  2. Get proposals and quotes from vendors
  3. contract start
  4. Introducing a recommendation engine to your EC site or website
  5. Make initial settings such as embedding recommendation tags and adjusting the layout
  6. Start operation, contact the vendor if you have any questions

Approximate time from vendor comparison to introduction is one month. You will need to go through procedures such as hearings and agreeing to the contract, so be sure to give yourself plenty of time to prepare.

Points to check when introducing a recommendation engine

What to check when introducing a recommendation engine

Review the following checklist before deploying a recommendation engine. Checking the checklist in advance will help prevent failures after installation.

  • Can it be expected to be cost effective?
  • Can it work with the tools you already use?
  • Is it compatible with smartphones?

Can it be expected to be cost effective?

In general, the higher the functionality of the recommendation engine, the more expensive it is. Compare multiple recommendation engines in advance, prioritize the functions you need, and then narrow down. If you choose a recommendation engine equipped with only the functions that your company requires, you can expect high cost-effectiveness.

Can it work with the tools you already use?

Some highly functional recommendation engines can be linked with various tools such as MA, BI, SNS, and data feeds. The wide range of cooperation with other tools is an important factor because it is a recommendation engine that requires user data.

Is it compatible with smartphones?

A recommendation engine that supports smartphones (multi-device support) can apply different designs for the PC site and the smartphone site. With more and more people accessing websites from mobile devices these days, mobile compatibility is essential.

Depending on the vendor, there are cases where the fees for PCs and smartphones are different, so be sure to check the fee structure in advance.

3 representative recommendation engines in Japan

There are many ASP-type recommendation engines, but for beginners, we recommend a vendor that has three points: ease of introduction, low cost, and abundant functions. From here, we will introduce a recommendation engine that has these three points.

NaviPlus Recommendation

It has been introduced to more than 500 sites, and is the number one recommendation engine with a track record of public introduction. It has the strength of being able to provide recommended content that reflects “action history”, “visitor leads”, “item attributes” and “visitor attributes” .

When implementing a system, they will propose the optimal implementation method and provide technical support according to the system environment and terminal environment of the site. Even after introduction, we are proposing improvements and utilization based on advanced cases, and the generous support is the appeal of this service.

RATE PLAN INITIAL COST: 200,000 YEN ~
MONTHLY FEE: 100,000 YEN ~
Main function Collaborative filtering
Visitor flow/attribute analysis Support before and after introduction of
automatic optimization function by AI AB test function

 

Rtoaster

A highly functional recommendation engine with a track record of over 350 companies. Many large companies such as Kuroneko Yamato and en Japan have introduced it. It features a wealth of functions such as data import, MA/BI linkage, and content analysis.

It picks up user data accumulated in CRM, SNS, websites, etc., and automatically creates segment information and target information using advanced machine learning. Furthermore, the range of cooperation is applied to touch points with users, and push notifications can be made using email and LINE.

PRICE PLAN (EXCLUDING TAX) FROM 200,000 YEN PER MONTH
*INDIVIDUAL ESTIMATE DEPENDING ON THE FUNCTION USED
Main function action+ (recommendation engine/user analysis)
insight+ (data integration/collection, etc.)
research+ (external communication such as LINE and email)

 

personalized recommender

The personalized recommender uses a unique recommendation system of “action history x product information x operator information” .

It features collaborative filtering that can propose appropriate products based on multiple user attributes, rule-based recommendations that can recommend specified products, and text mining that can immediately reflect recommendation information for new products with no action history. For products that are already popular to some extent, you can increase the number of sales by introducing a ranking function.

PRICE PLAN (EXCLUDING TAX) INITIAL COST: 200,000 YEN ~
MONTHLY USAGE FEE: 50,000 YEN / SITE
* PAY-AS-YOU-GO SYSTEM ACCORDING TO THE NUMBER OF ACCESSES
Main function Collaborative filtering
Rule-based recommendation Personalization
Recommendation
ranking function
Effect measurement function

 

Recommendation engine case study

Rtoaster

So, how much effect can we expect when we actually introduce a recommendation engine? We will verify the effect of the recommendation engine from the case of Hankyu Travel International and i-learning.

[Hankyu Travel International] Sales increased by 1.5 times with recommendations using core data

Hankyu Travel International has implemented a recommendation engine on the travel tour reservation and sales site Trapics. In cooperation with a DMP (data management platform) that accumulates core data, we are renovating the site so that we can propose optimal travel tours according to user preferences.

First, we started with browsing-based recommendations for one product across three devices: personal computers, smartphones, and mobile phones. After that, we gradually expanded the recommendation area. As a result, online sales have increased by 1.5 times, and the ratio of sales via recommendations within the site has improved to an average of 20-30% .

[i-learning] Succeeded in improving sales ratio and reducing costs through recommendations

i-learning, which supports corporate human resource development, has introduced recommendations in a learning video content distribution service called “Myra”. There are more than 200 types of learning courses in Myra, and we have made improvements so that you can search for the desired course more efficiently.

As a result, we succeeded in reducing the total cost by reducing the time and man-hours required for promotion. Furthermore, 20 to 30% of sales come from recommendations, which leads to improved sales efficiency.

Implement a recommendation engine to improve UX

A recommendation engine is an effective way to improve the UX of your EC site or website. Users can efficiently access the products and information they want, improving convenience. As a result, it will lead to an improvement in trust in the company, which will have a positive impact on business performance.

In order to effectively use the recommendation engine, it is important to check the cost-effectiveness and the range of cooperation with other services in advance, and then select the appropriate service that fits your budget and purpose . Also, check the scope of cooperation with external tools so that you can make use of your company’s core data.

Leave a Comment