Case Studies




Key questions




Increasing revenues and automatizing bids in online auction advertisement through Artificial Intelligence.

Discover how a leading company in digital marketing improved the ROI of online advertising campaigns through bid optimization.

Predicting and optimizing bids in online advertising auctions


A broker company for digital advertising wanted to automate a manually operated process, increasing revenues from their brokerage activities.
For the client, aHead Research proposed and realized a mathematical model able to suggest the best bid for gaining positioning over thousands of different keywords, so that better results could be obtained profit-wise from the auctions present on the platforms Bing and Google AdWords.
In particular, such optimization needed to identify the best offer for each keyword on each platform through predictions based on the analysis of data sets from the past, in order to maximize the expected ROI.

Key questions:

  • What is the optimal initial offer for a certain keyword?
  • Is it better to raise or to reduce the initial offer? Or is it better to keep the initial offer stable?
  • By how much should the offer be increased/reduced?
  • After how should an offer be increased/reduced? After 4 hours or after a full day?
  • Can the results obtained on a specific keyword be improved? Is it better to disable a keyword if it is not profitable enough?

If the country managers’ experience is enough to answer each of these questions individually, advanced techniques of Machine Learning and Mathematical Optimization can help them developing the business on a global scale. In fact, increasing the offer related to a keyword may increase traffic, but there is a chance that the revenues will not outperform the costs. Conversely, reducing the offer may lead to insufficient traffic and losses in online auction, which will also impact the revenues.





The problem

Increasing revenues by optimizing the Return On Investment of each keyword

Manually determining the best offer for each keyword prevents the business from scaling. In such cases, the support of a mathematical model comes to be necessary, so that an adequate offer can be identified. Such offer should not be so low that it becomes ineffective, or so high that it becomes too expensive. Through prediction, the use of Machine Learning for the study of historical data, and the development of a Mathematical Optimization model that factors in the earning power of each keyword, it was possible to design a solution that keeps track of past economic results to improve them in the future.

The solution

Our solution for bid optimization in online advertising

The teams of Data Intelligence and Math-Optimization of aHead Research developed a solution with algorithmic components organized in micro-services with REST API. The platform Apache Spark was employed to handle such a large volume of input data, so that an efficient data elaboration could be achieved. The software brings together a Machine Learning Model and sophisticated Python-based techniques of operative research. Scheduling the algorithmic workflow and showing the results to clients are tasks performed through Java, using stacks that increase the efficiency of such processes and identifying input data from an Amazon Web Services Cloud Bucket S3. These technologies made it possible to decrease the automation time of a process that would have required hours to handle hundreds of thousands of keywords. The software was impeccably integrated into the client’s platform, which can automatically import the results and upload them on Google or Microsoft.

The chosen algorithm can satisfy all of the client’s desired functions:

  • Evaluation of the bid on a daily and hourly base, with the possibility of changing the offer after a certain amount of hours, leaving freedom of choice to the user.
  • Simultaneous management of distinct markets at the global level, with different time zones.
  • Simulation o different scenarios and calculation of the KPIs to evaluate all possible scenarios.
  • Simultaneous management of different platforms, such as Microsoft Bing and Google AdWords.

The benefits

The benefits gained from bid optimization

The proposed solution allowed for the automation of computing processes and for the direct identification of the bid to place in the research engine platform. The software was able to suggest ideal offers with hourly granularity, so that the optimization could be near-real-time. Through the techniques of Machine Learning on which the software is based, it was also possible to identify clusters of unprofitable keywords, for which an immediate dismissal was suggested.
Apart from economically measurable results, automizing such process leads to a considerable saving of time with allows for an immediate improvement of the results. Mathematical optimization has therefore made it possible to obtain outstanding economic results, moving close to the profitability of the client’s marketing experts, consistently speeding up the process and allowing for a global scaling of the business. The benefits were demonstrated by collecting economic data before and after the optimization, showing how thousands of campaigns and hundreds of thousands of keywords could be handled at the same time and in several countries.