What if you could predict the performance of your ads with the same accuracy as the missile defense system predicts the path of the missile?
One of the speakers at the upcoming Pocket Gamer Connects in London will show all of us how advanced predictive intelligence technology is being implemented to optimise user acquisition campaigns.
Roy Leiser is the tech lead at yellowHEAD - a leading performance marketing company headquartered in Israel that focuses on organic and paid user acquisition solutions for growth-oriented mobile app and game developers.
As an expert in converting data into profitable insights with a passion for AI, Leiser will give a presentation called “From missile defense to user acquisition: How advanced AI models predict performance”.
Rise of the machines
Artificial intelligence technologies are being developed all over the world in different industries and fields. Nowadays, robots are working hand-in-hand with human experts, enabling them to do their job faster and better.
It’s therefore no wonder that user acquisition is now being powered by advanced computing tech just like the military, cyber-security and healthcare industries. In all these fields, laser precision and accuracy are highly important for the outcome.
In addition, the combination of infinite amounts of structured data from various sources, plenty of experienced and tech-oriented professionals, and significant marketing budgets made our industry the perfect playground for innovative AI systems and tools.
Earlier in September, yellowHEAD announced Alison, machine learning based technology that predicts campaign performance. Leiser was managing this project and took care of data preparation, model integration and validation testing.
He comes from a military background; for four years, Leiser served in the cyber-technology unit 8200 of the Israel Defense Forces, where he acquired his analytical skills.
“We wanted to equip our campaign managers with the capability of optimising faster and better and enable them to take smarter and more informative actions as early as possible in order to win the battle of wasted ad spend," says Leiser.
"The answer appeared quickly and with force when I compared Alison’s goal with my experience in the cyber-technology unit.
"We needed a technology that would exceed physiological limits of human campaign managers by computing huge amounts of data on the fly on one hand and at the same time execute regular tasks, allowing experts to focus on more strategic and complex missions.”
Powered by machine learning, Alison constantly learns and trains herself with new data. This is how she augments human analysis and decision-making by constantly capturing knowledge from active campaigns across different platforms and verticals, giving predictions with constantly increased accuracy.
Leiser will show an example of how one of the most advanced machine learning algorithms, called XGBoost, can be applied to a dataset of missile trajectory (speed, location, angle, etcetera.) with great efficiency. Similarly to how actual missile defense systems work in the battle field.
He will also go into how this model inspired the idea of building Alison using the same XGBoost algorithm.
Developed by top university math professors and yellowHEAD’s Data team, Alison calculates a campaign's probability of success based on millions of data points, predicting results across multiple ad platforms.
During the session, Roy will talk about the model and what challenges had to be overcome in order to successfully deliver substantial increase in ROAS for the world’s biggest app developers, such as Playtika, Scientific Games, GSN and many more.
He will go into technical details, sharing key learnings from the missile path detection technology and how it was used to achieve accuracy in predicting performance of paid user acquisition campaigns, a path that was both tedious and challenging, but quite impressive at the end.
Don’t miss Roy’s presentation that will take place at 12:00 pm on day one (January 22nd) in the Monetisation, Retain and Acquire track.