What happens when machine learning meets digital advertising?

David Renaudie

Data Scientist, Innovation & Research, Amadeus IT Group

Have you ever wondered how it’s possible that the advertisement on your screen shows exactly what you were looking for? Welcome to the wonderful world of digital advertising fuelled by machine learning. This makes many sceptical or afraid, but there’s no reason to be, we’re just entering a new era where technology enables relevant and respectful ad targeting.

phone as target

As a data scientist, there's nothing more exciting than playing with big data and fine-tuning top-notch algorithms to get concrete business results. It’s a fascinating process that allows us to make sense out of incredible complexity. This is my bread and butter, but to be honest, it’s even better when the computer itself makes sense of the data: this is called machine learning.

Machine Learning

Machine learning

is a field of computer science using artificial intelligence and advanced statistics to unveil patterns hidden in huge amounts of data. To do so, the computer automatically builds and updates a model by understanding the patterns detected in the data, allowing for better decision making. Predictive analytics is a good example of how using machine learning can help glean useful insights.

I've had a chance to apply these techniques to the digital advertising world. In this highly competitive market, being able to bid for the relevant ad space in a matter of milliseconds is key for generating leads and revenue. Ultimately, the objective is to purchase relevant display banner slots for advertising partners of travel audience, an Amadeus company specialised in digital advertising for travel. We worked hand in hand with travel audience data specialists to make this happen.

The Holy Grail is to be able to predict whether a user seeing an ad on a website will click on it, or not. If you knew in advance that a horse in a race has a high probability to win, then you would bet more on it, right? Similarly, a well calibrated click prediction engine enables the placement of smarter bids for ad slots. Machine learning helps make sense of past data in order to predict the future. But it doesn’t come that easily.

Applying predictive analytics to digital advertising

How we did it

First, you have to reach a deep business understanding of the data. It sounds obvious, but so many people rush into coding without having the slightest idea of the true meaning of the data they have. Once you start getting data insights, many questions will arise. The answers to these questions will uplift you well above average performance.

Second, you have to eat gigabytes of data for breakfast. Not literally of course, but manipulating this huge mass of information with ease is a must and technologies like Hadoop and Spark make this is possible. And as I said, this is just for breakfast.

Third, you have to pick the right algorithm. Ever looked for a needle in a haystack? This is harder. And let’s be clear, there’s no such thing as an ‘über algorithm’ that ‘suits all your-needs and brings you a coffee’. And you have to know each algorithms’ strength and weakness. Once you find the right algorithm, you pick relevant data-science metrics to evaluate your model’s performance, derived from the defined business goal. You play with the chosen algorithm, adapt it, and fine-tune it as if it were a mechanical Swiss watch. When that’s done, it begins crunching massive amounts of a data in a fraction of second. We can see how well it works with training data to predict future clicks – and yes, we can estimate this. It’s awesome, really.

Finally, how do we check whether all this stuff is really working? Our colleagues at travel audience plug the machine learning model in to one of the bidding platforms and watch it make decisions. Each KPI such as click-through rate and conversions are monitored closely. You watch the improvements with your very own eyes and see the click-through rates improve by not 1%, 5%, or 10% - they are multiplied by 3! And it’s only the beginning of how machine learning can lead to real business results.


Big Data, Artificial Intelligence