Adult Content Service Increased Ad Revenue by 10%

A Case Study in the Use of Artificial Intelligence

Adult Content Service Increased Ad Revenue by 10%
Портрет эксперта
Dmitry Shimko

Data Scientist at OrbitSoft

  • 01
    Client

    is a network that manages advertising on services with adult content — Agency.

  • 02
    The business objective

    is to reduce the number of refusals to watch ad videos.

  • 03
    The solution

    is to customize the display of videos, taking into account personal recommendations. For this, we used algorithms that allowed us to analyze and use user data.

  • 04
    Results

    +5% Increase in number of videos watched

    +19% Increase in viewing time of a single video

    +10% Increase in advertising revenue from videos watched

    -16% Decrease in refusals to watch ad videos

2 Billion Ad Impressions per Month

Agency is the largest advertising company in Canada. One of its projects is an adult content service. This is one of the most popular entertainment platforms: it has more than 40 million published videos and 350 thousand daily online users.

With the help of Agency, the service earns on ad impressions for their main content: for example, they use videos, banners, pop-ups, and other formats. At the beginning of 2021, the service had more than 2 billion ad impressions per month.

Agency manages ads for a popular service

  • >40 Million
    Published videos
  • 350k
    Users online every day
  • >2 Billion
    Ads impressions per month

The Problem Is That Users Don’t Watch Videos and Skip Ads

The adult content service makes money on advertising, their income is regular, and there are no problems with that. At the same time, there is high competition in the field of entertainment content: from game streams to videos with kittens.

When they contacted us, the average number of refusals on the service was 17.7%. This suggests that users were not that interested in the content offered. Because of this, advertising also suffered: the less time visitors spend on the service, the less revenue the service receives.

The problem isn’t just advertising revenue. It’s about attention: as soon as users stop spending time with you, your competitor gets their time. Returning such users to the fold is a long, expensive, and difficult process.

Limitations: There Is No Data for Personal Recommendations

The customer came with the following request: «The service collects a lot of different data. I’ve read that they can be used to increase content and advertising metrics. How should I do it?»

Our hypothesis: if you learn how to show an ad based on personal recommendations, then visitors will be more likely to watch the videos, which means they will see the advertisement.

The tool for setting up personal recommendations itself is well-known: for example, this is how YouTube, Spotify, and other services work. The problem is that we are talking about adult content, so there are differences from YouTube:

  • website visitors don’t share personal data — for example, their gender, age, country, and so on
  • they don’t leave likes or any indication that they liked one video or another
  • they use a VPN, so it’s impossible to know which country we are dealing with
  • at least 80% of visitors use incognito mode, and every time they visit, they are considered new users

The client faced the situation that he had lots of data, but he didn’t understand how to use it, how to customize the display of videos based on personal recommendations.

The Solution: Working with a Big Data and Creating Algorithms

Since the data is there, it can be processed in some way. To do this, we have created algorithms that allow us to collect and analyze it, and build personal recommendations based on it.

The project took almost a year and was put together in stages.

Project stages: creation of an algorithm for personal recommendations based on artificial intelligence

  • Data collection and analysis
    • Data collection
    • Analysis of found metrics
    • Pattern identification
  • Data preparation
    • Selection of meaningful metrics
    • Conversion of data to desired format
  • Modeling
    • Choosing a forecasting model
    • Experimentation with test datasets
    • Development of a recommendation algorithm
  • Review of recommendations
    • A / B tests
    • Comparison of results
    • Analysis of gained experience
    • Planning further steps to improve working model

What Helped Solve the Client’s Problem

Accumulated experience in the field of working with big data. This is not our first project in this area. For large clients, we process billions of lines of user behavior data. Thanks to this data, we know how to:

  • calculate the ad rating and display it in accordance with the rating
  • determine the likelihood of a click on an ad
  • match the user with a specific segment of interests to display relevant ads

Thanks to this accumulated knowledge, we already knew which models and algorithms would be suitable for the service. For the customer, this means that we can make their project faster and more accurate.

We didn’t waste time inventing a solution from scratch. Instead, we used working solutions from areas close to the client’s business. For example, we studied documented examples from Google, Netflix, and Spotify. Our research helped us to quickly determine the best methods and approaches for our own development.

A page about algorithms from a document from Google

Left side — a page from a document from Google employees discussing recommendation system algorithms for YouTube. See the entire document in PDF

Right side — Google’s experience and mathematical models in building scalable recommendation systems. See the entire document in PDF

The most interesting reports were printed and analyzed by the whole team

The Result: Personalized Recommendations Increased Ad Revenue by 10%

Based on almost anonymous data, we created an algorithm that selects videos based on personal recommendations. Thanks to this, the service has increased the main indicators of advertising efficiency. Another project for Agency.

1.5 months results: how AI-powered recommendation system influenced content viewing rates. Results how AI-powered recommendation system influenced content viewing rates

  • +5%
    Increase in number of videos watched
  • +19%
    Increase in viewing time of a single video
  • +10%
    Increase in advertising revenue from videos watched
  • -16%
    Decrease in refusals to watch ad videos

Earn more on advertising — Subscribe to our newsletter

Where else is the algorithm for developing personalized recommendations useful?

The algorithm is suitable for any online business with a large amount of content and number of visitors.

Content is anything like:

  • articles
  • videos
  • audio
  • memes
  • cards with goods or services

If you have a similar business, write to us and let’s discuss how we can help you, too.

Whatever your needs, we can help!

Tell us what problems you’re facing with your business. We look forward to hearing from you.

Получите ответ по смс

Ваше сообщение успешно отправлено!
Представьтесь пожалуйста
Укажите номер, на который придет ответ
Нажимая на кнопку, вы даете согласие
на обработку персональных данных.