Skip to main content



In the world of digital publishing, we are always looking for solutions without effort, with high added value.
The industry is saturated with stories from publishers who have modified a simple element, such as the location of a menu or its appearance for example, thus increasing their revenues and at the same time improving the visitor experience.
And if there was a solution to bring more value added than any other known solution?

This is precisely the subject that Iggy Chen, director of the business intelligence division at Ezoic, addressed in his recent case study at Google in New York during the Pubtelligence.
Iggy emphasized the considerable value that the segmentation of visitors to a site brings to digital publishing . He showed how simple differences in visitor processing can have a huge impact on long-term revenue increases.
Below, I will publish the results of his research and examples  and how to implement this on your sites.


All visitors to a site behave differently. Their behavior and willingness to browse the content of a site directly influence the income reported by the site.
Even without a direct agreement with the advertiser, the publisher's earnings are affected by the behavior of his audience . In programmatic exchanges, visitor engagement is directly related to the amount that advertisers are willing to bid (or not) for the publisher's inventory.
In addition, the way a visitor browses the pages of a publisher's site directly affects the number of ads displayed per page, which also affects the EPMV  (revenue per thousand visitors or revenue per session).

We will not give up, however, because the behavior or engagement of the audience is not impossible to change and predict. 
We all know that changes to a site's pages can directly affect the number of pages a user visits, the time they spend on the site, their willingness to come back, and so on.

Iggy said it was like comparing fantasy football league groups within the company. Football fans joined the fantasy league unlike others who created their own league.
Both leagues had a very different behavior in the way they followed the season.

The purpose of the Iggy case study is to determine how it is possible to categorize visitor behavior of a site similarly.
What it shows is that the essential thing is not to find  a significant and unique change for all, but rather to identify the minor changes that could correspond to each type of visitor and thus improve certain parameters such as the number of consultations. of page per visit, the duration of engagement as well as the bounce rates.
Discover here the whole Iggy presentation.


The case study presented at Google comes from the TigerNet publisher .
Tigernet is an online community dedicated to supporters of Clemson University . This community has a large contingent of loyal readers, but also many, many more casual readers from the university sports community.

Just as Iggy was able to compare the different behaviors of participants in a fantasy football league between colleagues, he explains that the segments that are Superfans and Casual Users are present on all sites, including Tigernet.
He admitted, of course, that the segments are in the thousands , but he explained that it was better for this study to focus on two fundamentally different user groups to discover the significant financial benefits. specific treatment of each of them.


First, let's look at how the case study was done.
TigerNet uses Ezoic's machine learning platform to learn more about its visitors. The platform integrates machines that memorize the visitor's behavior over time and defines certain elements such as location, type and density of advertisements to maximize the engagement and duration of the session.

In this example, we will analyze two types of specific visitors that Ezoic has identified over the years. Iggy uses the terms he has referred to previously, Superfans and Casual Users .
We can consult some of the attributes of the Superfans above. They often come from the state in which Clemson is located, often do not visit the site for the first time, often visit it on mobile and frequently visit it.

The study looks at how the machines treated these users differently based on their previous reactions and many variables.
The old articles are at the bottom of the page because it is the home page of TigerNet. The machines chose to place two ads around the Superfans' main point of interest, the top of the page.
It can be assumed that the machines have determined that the Superfans are familiar with the old news broadcast on Clemson, and that the ads embedded in this site have a better success rate.
As has been said before, placing ads in places where the engagement of the visitors is the highest often makes it possible to make more profitable the advertisements ... It must nevertheless be put in relation with other elements such as that the ability of these advertisements to increase the bounce rate and shorten the duration of the sessions .

When the Superfan moves from the home page to the next page, the machine only displays two additional advertisements for a total of four on the session.
Previous data seemed to indicate that the maximum threshold for getting more page hits for this type of visitor is about two ads. Machines have maintained an ad density below this threshold to ensure that the session does not end earlier than usual.

Finally, you'll notice that this last ad is at the very end of the article. Since the  previous data showed that the user will probably want to read the entire article , the machines place the high-value article at the very end of the content instead of putting it in the middle at the risk of interrupting the article. his reading, especially since he is unlikely to do so to click on an advertisement.
Fewer items are ignored or considered non-searchable, which keeps a high advertising value on the site. This helps to maintain a higher competition on the site for the advertising space of this page while maximizing the value of this individual session.
Understanding this user's behavior allowed the machines to evenly distribute the ads throughout the session, increasing revenue and improving the user experience.
It also kept the ads in places to avoid dilution .


Now, let's look at how machines have used previous information about visitor behavior to treat another segment of visitors differently.

In this case, we are talking about the  occasional user . Machines have learned that visitors with these attributes (not from the Clemson state, from a social network or surfing at night) typically spend less time per session.
Since it is rare for these visitors to view more than one or two pages, the content is monetized differently.
Since these visitors generally browse the site globally and rarely go down the page, machines place high-value ads at the top of the article's pages where they are most likely to be seen. .
This approach allows machines to maintain an EPMV (total revenue per session) higher than it would have been with static ad placements presented to Superfans.
By treating these two users differently, the machines were able to maintain a total EPMV for both user sessions approximately 45 percent higher than it would have been if both users had viewed the same density and placement of ads.


In this case, one of the ways that machine learning works is that the machines themselves begin to determine the most important factors for advertising value and user experience on each site.

In this case, we have identified several interesting factors.
These two visitors are very different and the machines have learned to recognize the attributes to categorize them.

This involves monitoring and understanding these indicators for different visitors:
These indicators provide a more granular level of data regarding visitor engagement with site content .
This allows publishers to gain a long-term view of monetization and maximize return on investment.

You can find this information for free in the new Ezoic Big Data Analytics Suite, which publishers can access at no cost.


Beyond the ability to segment visitors and treat them differently, Iggy's case study also shows real-world earnings data in real time.

In one example, he showed how machines can quickly adapt to a new influx of visitors.
The increase in visitors was directly related to Clemson star striker Greg Huegel's injury.

During the analysis of the data, the machines were able to realize the change in the behavior of the users during this extraordinary event.
Many users from social networks have flocked to share news on Facebook, Twitter and Reddit.
These users have never visited the site before or do not visit it often.

If we look again at Superfans and Occasional Users, we can see that there are big differences in the way visitors are treated for this scenario and that some scenarios are more optimal than others.
In the case of the SuperFan, the machines learned that it was likely that he would continue to browse the site; as a result, they have disseminated ads throughout the user's journey, helping to achieve consistent CPM rates at the page level.
For the Occasional User,  the machines chose to display more ads on the page as they determined that they were more likely to only view this page and leave.
If we take a closer look at the case of the Superfan browsing the homepage rather than the article page, the machines will choose to display some advertisements because they know he knows the layout of the site and that the fact of having to view advertisements to find the content he is looking for does not bother him.
When we compare  these two types of users, we find that they have very different optimal EPMVs . The publisher should not only guess what type of visitor it is but also display the optimal pages to maximize its EPMV so that the publisher does not lose money.
This is why it is essential to have machines that can learn and filter this data.


Iggy Chen states that each site has its own version of these two types of visitor segments. It also indicates that there are probably thousands of other segments of visitors.

He discussed how visitors can be grouped into two different types of data.
Elements that publishers can impact:
  • Navigation options present (sidebar, menu, links)
  • Page Loading Speed ​​- Critical Path Determination
  • Content positioning
  • Types of advertising (display, native, link, video)
  • Location of ads (location on the page)
  • Sizes of advertisements
  • Presence of other advertisements and their location
Elements that editors can not control but must consider:
  • Type of operating system or device
  • Browser type
  • Time of day / day of the week
  • User connection speed
  • Geographical location
  • Source of upstream traffic
  • User experience indicators from other visitors to the page
  • Subscriber or non-subscriber (possibly)
  • Real user history
  • And much more
Combining this data will allow you to start separating the different segments of visitors. According to Chen, each of them will behave differently.


If we stay on our TigerNet case study , we can see a significant increase in revenue, improved user experience indicators as well as more pageviews .

Through the identification and tracking of user behavior, the case study showed that publishers can use this technology to optimize decision-making to improve user experiences and increase the total value of the user. 'user.
This is a new element for publishers; just like hand-tuning, these variables would be excellent but can not be scaled. This has the effect of providing an optimal and similar approach to publishers and advertising operations professionals.
In the end, treating users appropriately makes them happy in the long run, and happy users involve happy publishers.


Want to try on your site what has been implemented on TigerNet? Ezoic is a free platform for publishers to try these things themselves.


Popular posts from this blog

How To Create A Blog

How To Create A Blog: "Our 9 step tutorial will teach you how to create a blog in 20 minutes or less! Learn about the best blogging platforms, getting a $.99 cent .com domain name, choosing a stunning blog theme, and most important things to get started profitably!  I started my first blog almost 9 years ago and the journey has been life-changing for me in dozens of ways. I've made tens of thousands of dollars, lived in many tropical countries (currently Thailand), met hundreds of amazing entrepreneurs and much more. When I first got started it was simply for fun and some side income. I never expected it to turn into such a high-paying career. But, I'm not special - I'm just like you, an average guy. However, years ago I developed a system of how to do keyword research and know how competitive it was to rank in Google. After analyzing the competition and know if it would rankor no built built dozens of different affiliate blogs with the exact process I am sharing below.

How to Make a Website With Your Own Domain For Free

How to Make a Website With Your Own Domain For Free Today, we’re going to talk about domain names. ‘Do website builders allow to connect custom domains?’ is a common question we regularly get via our e-mail. Other popular domains-related questions are ‘Can I use a previously purchased domain name’ and ‘Are there any site builders allowing to connect my own domain for free?’. The answer to all these questions is yes.
Do website builders allow to connect custom domains? Usually, when you sign up to a website builder and start working on your website, you get a free subdomain which is a temporary address of your site while it’s under construction. With Weebly, for example, you’ll get an address like sitename.weebly-dot-com, while with Wix it will be formatted the following way: yourlogin.wix-dot-com/sitename. During the setup stage you can use a free subdomain, but when your website is ready for publishing, you can easily attach a new or a previously purchased domain name. We’ve already …