How Machine Learning is Optimising Spending and Eliminating Fraud - Impact Radius

Tim Maytom - Sponsored by: Impact Radius

David A. Yovanno, CEO of Impact Radius, explores how the rise of machine learning won't just help marketers spend their money in the most efficient ways, it will also help root out fraud, as long as it's paired with human expertise and judgement.

“Artificial intelligence” and “machine learning” are two of the hottest buzzwords in the industry right now, but many professionals in our space are still working out what these technologies do and how they can be used to improve digital marketing.

This week, Oracle announced implementation of AI-driven capabilities across its Customer Experience Cloud. SalesForce recently launched joint artificial intelligence solutions in partnership with IBM’s Watson AI technology and others aren’t far behind as understanding and demand accelerates.

AI is already being leveraged by advertisers to do anything from creating chatbots to crunching massive data sets to drive hyper-targeted media buys. As CMOs explore new ways of improving return on digital ad spend, they must stay informed on the key attributes of this emerging technology as a proven solution to make our ecosystem more pure. Here are the finer points of artificial intelligence and machine learning necessary to understand how brands can leverage these tools to enhance their targeting and weed out fraudulent impressions.

So, what is the difference between artificial intelligence and machine learning?
Stanford’s computer science department defines artificial intelligence as “the science and engineering of making intelligent machines, especially intelligent computer programs.” In this context intelligence is the “computational” aspect of how we humans achieve our goals in the world.

Machine learning is a subset of AI that focuses on creating machines that “learn” without being explicitly programmed to do so. At their core, machine learning algorithms are designed to solve optimization problems, often to make some sort of prediction, such as how do I accurately predict which consumers have a high intent to purchase.

While machine learning was largely out of reach merely a decade ago, the rise of increasingly powerful CPUs to process the data and an almost infinite amount of cheap data storage has given us the power to parse previously unthinkable loads of information. As you might imagine, this ability is one that’s having a major impact on how brands and agencies approach digital advertising.

Machine learning helps brands buy the right impressions with incredible precision
Machine learning allows brands to take targeting to the next level by using algorithms to first analyze user behavior and then make predictions of which users are the “best”, which can be defined as highest intent to purchase or LTV, most likely to engage, or many other KPIs.

For example, a machine learning algorithm might process a training set of the site data of prior purchasers of widgets and then pinpoint a precise group of users as likely customers based on a range of detailed information, including how long the users spent reading widget-related content, what percentage of their browsing activity was devoted to widget review sites, whether they have interacted with an ad before, the type of interaction and so on. Already, marketing agencies like Publicis.Sapient are using these tools to automate more mundane functions to become more effective and efficient at their jobs and able to focus on more high-level work like strategy in order to achieve better results.

Machine learning is conquering the bots -- and who knows what’s next
One of the most innovative applications of machine learning in marketing is bot detection. Bad actors are stealing billions of dollars from brands each year by using bots to generate fake impressions and clicks. These bots can range in complexity from General Invalid Traffic, such as automated browsers running within data centers, to Sophisticated Invalid Traffic, such as bots that copy user mouse movement and accurately simulate human behavior.

A common approach to detecting bots has been rules-based analysis. For instance, a simple rule could be flagging users that generate above a certain number of clicks each day. However, the flaw with rules based approaches are that black hats can always re-work their tacts to be just below the thresholds that would be flagged. Machine learning allows us to take a more sophisticated approach to fraud detection by analyzing how users browse with the internet and calling out anomalous behavior, without a necessarily pre-defined notion of what “anomalous” really is.

The goal of a bot on the internet is to steal ad revenue, whereas a human has much more benign goals, such as consuming content. Because of this bots have to browse in a way that is inherently non-human, and this anomalous behavior can be called out by learning algorithms can segregate the human from non-human activity.

Machine learning allows us to Identify and track nonhuman behavior
While machine learning allows us to identify groups of users which are browsing similarly, it takes additional human analysis to identify which patterns are anomalous and associated with nonhuman behavior. By combining this pattern recognition with a team of human experts capable of identifying bot behavior, ad-tech vendors are able to help brands and agencies catch on to fraudsters’ latest schemes in real time. In fact, this mixture of machine learning behavior clustering and human intelligence is how Forensiq discovered that brands had lost nearly $1 billion to black hats who were hijacking people’s phones to continuously run ads they couldn’t see.

Moving forward, the industry will continue to find innovative applications of machine learning technology. In the context of fraud detection, this is especially important as fraudsters continue to become more sophisticated and better able to hide in the noise of the vast amount of traffic on the internet.

By taking proactive steps to better acquaint themselves with this emerging technology, brand and agency-side marketers will put themselves in a position to make the most of whatever new innovations come down the pike. While the intelligence may be artificial, machine learning’s potential to transform digital advertising is very, very real.