Get In Touch
Please feel free to contact with any questions using the form below
By clicking on the button, you consent to the processing of personal data and agree to the privacy policy
Email Marketing | Analytics
Product Recommendations in Email-Campaigns: How It Works
Tools of a modern email-marketer are becoming more and more diverse and accurate. Many companies track and analyze the product preferences of their clients and then place product recommendations on their websites and in email campaigns. This is understandable, because the market demands relevant and personalised content for e-commerce specialists.

When we say "market" we mainly mean the clients themselves. According to a survey conducted by Harris Interactive and Listakr companies 84% of customers - subscribers of promo campaigns - find them necessary and useful if the offers correspond to their preferences. 64% of them are ready to tell the seller about these preferences to get more personalised offers.

Today we will tell you about sales tool called product recommendations in email campaigns.

It suits businesses with more or less large catalogues of products. They may be online stores for clothes, cosmetics, sporting goods, electronics, household appliances and other things.

Types of recommendations in email campaigns

We do not take into consideration simple methods: to show accidental products, mostly or earlier viewed products, highest margin rate products, products from a dropped basket, etc. We will take into consideration similar and related products.

When similar products should be shown:

on the description page of the product to facilitate the choice;
if the customer subscribed for the arrival of the product which is out of stock now - maybe he will buy what is in stock now;
in reactivation letters if you suppose that the product bought has been already consumed and the customer may want to buy it once again.

When related products should be shown:

if the customer added a product to the basket and placed an order - make sure he doesn't forget about accessories and component parts;
in reactivation emails after placing an order - let him buy accessories, spare and component parts.

Later we will discuss the kinds of recommendations - manual and automatic ones.

How manual recommendations work in email campaigns

The easiest way is to add recommendations manually, like, for instance, DeAgostini publishing company does. In the email campaign dedicated to a model kit of "Sevastopol" battleship they offer a subscriber instruments for modeling. Products for the campaign were chosen by a manager:
Addition of product categories,for example, best-sellers and novelties, may also be attributed to manual recommendations.
Seasonal products may also be recommended. For example, FC "Zenit" included offers with branded school kits into their August and September campaigns:

Email Automated series

Product recommendations which are much more effective and less labor-intensive are automated ones. When we say "automated mechanics" we mean software solutions which track and analyze actions of users on the website: page views, interest to some groups and categories of products, filling in forms, adding products to the basket, tabs, purchase history, age, gender, preferences. The result of this automated analytics is a personalized dynamic code with the products which are most likely interesting for your subscribers.

The advantage of automated series is that you can set up emails with recommendations once and then they will sell for you 24/7.

Here is a list of emails which should include product recommendations:

Promotional emails;
Welcome emails;
Transactional emails;
Abandoned views;
Abandoned web-forms;
Abandoned cart;
Product in stock;
Price-drop notification;
Thank-you emails;
Delivery notification emails,
Any service emails.

How the algorithm of automated product recommendations works
The choice of algorithm depends on the stage on which you want to recommend products to the customer.

If a customer is only choosing the product, he should be offered similar goods.

If a customer has already bought the product, he should be offered related goods. Many stores offer goods from the same category as the ones in the basket or the ones with similar characteristics, but this way we offer the same goods again. It is necessary to apply more difficult algorithms to recommend exactly related goods.

In 1994 employees of IBM Rakesh Agrawal and Ramakrishnan Srikant published a scientific article «Fast Algorithms for Mining Association Rules» in which they described the algorithm called Apriori. It is a self-learning algorithm which traces order history (basket lists) and calculates product recommendations for further purchases.

Later, the algorithm was modified for faster work and giving recommendations based on marginality of the product, thus made it possible to offer goods which are more profitable from the seller's perspective.

The Apriori Algorithm is one of the first algorithms of artificial intelligence. It gives very accurate recommendations but it needs lots of data for learning.

If you have just opened an online store and you don't have any order history, you will have to use simple mechanics for product recommendations. If you want to recommend goods taking into account gender, you need to have twice as much teaching data: one set for each gender. Otherwise, either the recommendations will not be precise or Apriori algorithm will not be able to recommend anything.

So, if a CRM-platform makes product recommendations but does not require order history and does not collect information about previous orders, it indirectly means that its methods of making product recommendations are far from artificial intelligence.
Alexander Toporkov, web-developer

Platforms that work with product recommendations

There are a lot of both CRM-platforms with the function of product recommendations and separate recommendation services. We will describe only the most interesting ones, in our view.

Exponea

The mechanism of Exponea is, probably, one of the most comprehensive and interesting. On the web page there is a script installed which allows you to analyze the behaviour of a user. The platform allows to track all of the actions of a client, including, for example, offline purchases.

The recommendations themselves are based on the catalogues and can use any elements from them: products, promotions, articles and others. They are based on the four elements:

  1. The catalogue, which is the basis;
  2. The principle of choosing elements from the catalogue;
  3. Blacklist of the elements which should not be chosen (for example, goods which have already been purchased);
  4. Recommendation ranking principle (also based on the behavior of a user, for example, goods from a favourite category will be shown higher).
Then these recommendations can be shown on a webpage or inserted into any communication including email-campaigns:
Actions of the user with recommended goods will appear in his profile in the form of events. On the basis of this it is possible to make a report about the effectiveness of recommendations.

There is also a function of A/B tests in the service which allows to test emails with different blocks of recommendations.

ExpertSender

The platform has a special self-learning module to work with recommendations. When the script is installed on the website, it analyzes actions of users. The script needs some time for "learning" to give the correct data. The amount of time depends on the quantity and quality of traffic, one or two weeks on average. As a result we get a personalised dynamic code and add it to email-campaigns.

Mindbox

The function of product recommendations is realized in several ways:

with the help of the integration with services REES46 and RichRelevance when you just upload the list of goods manually;
with the help of algorithms which calculate similar, popular and related goods.

What is possible when you work with Mindbox algorithms:

Different ways of setting algorithms - there can be three or five variants of one recommendation.
The algorithm of similar goods has a setting for recommendation display priority.
The self-learning algorithm of related goods. Its functions can be broadened to choose from certain products.
which categorie recommendations shall be chosen.
Algorithms automatically exclude goods which have already been purchased by a client.

Retailrocket

This is one of the first Russian platforms which initiated personalised product recommendations.

It makes pretty accurate product associations even if you do not have an order history. You export your product catalogue into the platform. Then the platform defines by name and a set number what product it is and on the basis of its data, collected over years of work, makes product recommendations.