Homepage > Portfolio > Decision-support system optimization for advertising industry

Decision-Support System Optimization For Advertising Industry

Improving performance of a system that is designed to compile best offerings for advertisers to broadcast their commercials. Tackling the problem of missing data through applying a data approximation technique.


Distributing information to a large number of users in order to increase your own target audience and find potential customers is an important part of each company.

Before the American entertainment company decided to contact Azati, it prepared a clear request and the necessary data. As any successful and fast-growing company, it collects and structures data from its customers in order to analyze their needs and preferences.

At the time of contacting Azati, the company had collected information on its TV viewers, who are also divided according to certain characteristics and categories. Since television is popular with a lot of people, it was quite difficult to distribute them all according to each characteristic, so they generalized a bit.


Ads are everywhere. You turn on the TV, open up an internet browser, or walk down a street and you’ll see ads. They’re all clamoring for your attention and – ultimately – your money. Ads can be annoying, so we have to take into account the target audience.

Knowing your audience helps you figure out what content and messages people care about. Once you have an idea of what to say, knowing your audience also tells you the appropriate tone and voice for your message. So the customer turned to Azati to create a decision-support system.

With the help of such a decision support system, companies intending to place relevant information for this or that audience at the right time for the consumer can be almost 100% sure that their advertising campaign through television will be able to interest the potential consumer who will be ready to buy the product. .

The main task was to reconstruct all the processes into a continuous dependence, using the known values of the characteristics at discrete points in time for the full performance of the system.


The advertisers want their adverts seen ideally by specific audiences that possess certain characteristics.

For this, the client company gathers information on as many as ~60’000 TV viewers that are described by ~30,000 different characteristics. By integrating, transforming and analyzing this data, the elaborated system is able to propose optimal broadcast plans to advertisers.

The optimal plans are best options found at the intersection of the following requirements:

  • Characteristics of the people the advertiser is willing to target;
  • Number of targeted audience views;
  • Budget an advertiser wants to pay for the service;
  • Duration of the commercials and more.

Importantly, the system also envisages and indicates the perfect TV time for an advertiser to place their advert, so that it is watched primarily by the people they are willing to target.

To make this happen, the system finds out most likely viewer characteristics intrinsic to the viewers of the particular program at a particular time in the future.

In order to make calculations of the optimal plan for any given point of time possible, it is necessary that the system contains the information on each characteristic at each point of time. However, the system contained only partial data, as the data was uploaded with some periodicity.

The major challenge was to restore a continuous dependence, using the known values ​​of the characteristics at discrete points of time.


We analyzed the data for several periods of time. Then, with the use of linear regression with L2-regularization we were able to achieve approximate data with appropriate accuracy.

Therefore, the system became able to offer optimal plans for advertisers at any given piece of time.



Drop us a line

If you are interested in the development of a custom solution — send us the message and we'll schedule a talk about it.