Optimization of Online Advertisements

Optimization of Online Advertisements

Have you ever noticed websites present you advertisements for sunny holiday destinations when it’s raining outside? Or offer insurances when there is a crisis? If you haven’t yet, then you most likely will in the future. This optimization of ads displayed on websites is at the heart of Adtrackxys, a startup company whose technology we helped build up. Amazon has proven that recommending the right products to the right people is worth billions of dollars.

The basics

Advertisements shown on websites generate money for the website owners. The ads shown often come from large ad servers (e.g., OpenX) and the revenue from advertisements can come in different forms.

CPM (Cost Per Mille) CPM is a famous revenue model, where the website owner is paid a certain sum of money for every thousand impressions of an ad on the website. So the more visitors, the more revenue. This model is less interesting for advertisers as they end up paying for ads displayed to uninterested website visitors, thus wasting marketing budget while at the same time making the websites less coherent.
CPC (Cost Per Click) and CPA (Cost Per Action) is a revenue model where a website owner gets paid every time a website visitor clicks on an ad on the website. Under the CPA model the website owner gets paid every time a website visitor clicks on an ad and then performs a certain action, e.g., a registering or buying an item. If the ad is well matched to the content (and thus the website visitor), then this can result in more ad income for the website owner as well as better budget spending for the advertiser.

The tendency is for companies to lean more and more towards the result-driven CPC and CPA models.

Let’s look at some numbers. Imagine a website that has 300,000 monthly visitors and that receives $0.50 for each advertisement that has been clicked on. If 5% of the visitors click on an advertisement then the monthly ad revenue is 300,000 * 5% * 0.50 = $7,500. If optimization of the advertisements can bring up the click rate from 5% to 6%, then the revenue goes up to 300,000* 6% * 0.50 = $9,000. In other words, there’s financial gains in optimizing the advertisements shown. This is exactly what Adtrackxys does. The algorithms at Adtrackxys pull together any information it can on the user (e.g, reverse IP lookup) and on his/her environment and responds in real-time to changing click behavior of the website visitors (e.g., show more advertisements for sunny holiday destinations when it’s raining).

The business model of Adtrackxys is interesting and worth mentioning. Adtrackxys gains money by receiving a percentage of the revenue increase due to the optimizing the advertisements. If there is no optimization, then Adtrackxys earns nothing, and the better the optimization, the more they also earn. It’s a win-win situation for everyone where not optimizing the advertisement is the worst a website owner could do.

What Mathecsys did for Adtrackxys

The optimization algorithms were developed by Ger Koole at the Vrije Universiteit in Amsterdam. As is very common, there was a lot of work that had to be done to take the algorithms from a research and experimentation prototype to a fully operational, robust, and well documented client-ready system. To accomplish this, the mathematical concepts had to be understood, algorithms had to be enhanced and created from scratch, and projects had to be well structured.

In particular, Robin Groenevelt played a core role in designing, creating, and implementing their technical needs :
– Creation and enhancement of optimization and forecasting algorithms
– Data analysis and defining data needs
– Build a UI front-end in Flex for visual data dependencies
– Helped migrate and modulate the system from C++ to Java

Leave a Comment