How To Fingerprint Ad Fraud Using Artificial Intelligence
ADOTAS – Digital ad dollars are expected to reach $50 billion in the U.S., and spending on real-time bidding is expected to increase 43.4 percent in 2014. But with the recent study from the Interactive Advertising Bureau (IAB) stating that 36 percent of all Web traffic is fraudulent, advertisers are now questioning their digital investments. Brand and performance advertisers suffer from this the most. This is due in large part to the lack of transparency needed to ensure a campaign’s performance is working against fraud.
So, how can advertisers fight fraud to ensure each campaign meets key performance indicators? They must first understand programmatic advertising, where a large portion of fraud occurs. According to a recent Forrester survey, only 23 percent of marketers understand programmatic buying. Programmatic ad buying enables marketers to let machines do the work for them, instead of manually combing through mass amounts of data to negotiate the best campaign possibilities. These traditional human practices are both inefficient and costly.
Fraud is a multi-million dollar business, and fraudsters have incentives to continue and get better at it. The problem with fraud is that is that each act is committed on a small scale, so it’s difficult for an advertiser to manually identify every strange behavior as fraudulent. This is where artificial intelligence and machine learning can help. By analyzing big data, the machine identifies and pieces together the fraud sending traffic from each location, and eliminates it immediately.
It’s not enough to just create basic fraud fighting rules, as experts are constantly changing their patterns. In order to succeed in fighting against them, the machine must evolve and learn from past experiences. The machine is continually self-improving, identifying its own rules and definitions, and applying it across multiple campaigns to minimize fraud in real time. The machine views traffic ratios between impressions, clicks, conversions and the time between every action, based on past experiences of a specific advertiser or a specific property. It can also detect the behavior from IP addresses. For example, if one IP address is registering to a service a hundred times a day, the machine is likely to flag the address as fraudulent sending traffic. The self-learning technology automates the decision making based on relevant customer data, ensuring campaigns quickly respond to market dynamics, cutting out elements that aren’t working, and capitalizing on the ones that do. This enables advertisers to reach audiences on a much larger scale, while also reducing overhead costs.
Manually optimizing campaigns, on the other hand, will enable the same mistakes to be made over and over again – especially for advertisers that don’t understand programmatic buying and where their mistakes are being made. However, the human input is still extremely important. If advertisers do identify strange campaign behaviors, they should cooperate with their trading desks, agencies, networks and tech providers to investigate the behaviors and blacklist the sites, if needed. It’s similar to your email provider. Your provider continues to improve on identifying spam, but every once in a while, you might receive a spam email in your inbox. But after you click, “report email as spam” you help your email provider learn, adapt and finally blacklist the spam sender.
The machines are not made to replace advertisers or contradict their decisions, but simply enhance the decision-making process. With the combination of human campaign rules and machine definition, advertisers can quickly identify ad fraud, keep pace with key performance indicators and run campaigns that generate the highest conversion rates. To improve campaign performance, cut cost and also fight fraud, advertisers need the support of machines.