Inuvo Details New Fraud Prevention Features

In my last post, I mentioned several projects in the pipeline for Inuvo’s analytics team. One project is bringing transaction-level cost-per-action (CPA) fraud detection to the Inuvo Platform.  Our goal is to create a system that automatically inspects every CPA transaction handled by the Platform, and determine the likelihood that it’s fraudulent.  That information could then be used in a variety of ways – from automatically denying extremely high-risk transactions, to creating reports that will let us tell at a glance when there’s a change in a publisher’s level of traffic quality.  This will make life easier for everyone.  Advertisers utilizing the Platform can be alerted to suspicious activity sooner, Inuvo account managers can spend more time interacting with clients and less time poring over transaction logs, and even publishers can benefit from the increased confidence that advertisers have in the marketplace (which translates into better payouts!).

We recently implemented the first version of this system and we’re excited about the preliminary results.  We’ve already rooted out several fraudsters that had previously escaped detection.  Some had remained hidden because they were extremely low volume, and it’s hard for a human to notice patterns in small amounts of data. Some hid their fake transactions among a larger volume of legitimate traffic, which again masked it from human review. Good news is that we now can detect these cases – we recently shut down a number of bad publishers after as few as just four transactions! Since we are scanning the entire Platform, in real time, we see much more traffic than any given advertiser on the Platform.  So when we shut down one bad publisher, every advertiser benefits.

The CPA fraud detection system (which we don’t have an official name for – suggestions are welcome!) learns from historical data, and makes distinctions between known fraud and known good traffic patterns.  This way the system can tell the difference between good and bad traffic. The only way to get past it is to be just like good traffic, which is basically impossible to do unless you are good traffic!

Of course, humans are also quite good at pattern recognition – when there are few variables, and they can be represented graphically, we can compete with computers.  So, besides giving the system raw data from the Inuvo Platform, we also give it “hints” in the form of custom variables designed to correlate with patterns that we’ve noticed ourselves, or that have been pointed out to us by our great Inuvo account managers.  The result is a system that combines human expertise with a computer’s ability to sort through mounds and mounds of data. The computer algorithms we’re using have proven themselves in shutting down fraud in other parts of the economy, too, so we feel like we’ve brought a big stick to bear against this problem.

The best part is that it will only get better from here.  The system works by making associations – by learning.  As more and more transactions flow through the Inuvo Platform, and as we get more feedback from advertisers on potential fraud, the system gets smarter.   

If you’re an advertiser or network and you’re having issues with handling fraud in a scalable way, contact our sales team at /Users/alex/Documents/sales [at] inuvo [dot] com">sales [at] inuvo [dot] com to discuss how we can help you.  

Oh, and, if you’re a bad publisher ... I hope you have a good lawyer! :)