Media quality is influenced by several variables, some more easily measured than others. Perhaps the most difficult to accurately measure is also considered the most ominous: fraud. As an industry, we’ve achieved acceptance (fraud exists) before understanding (how we measure it and how much is out there). In come the verification vendors, touting the ability to report what everyone is up against, except that their numbers don’t match up. Fraud levels are reported from as low as 2 percent to as high as 40 percent, with winners and losers at both ends of the spectrum. With such disparity, it’s no wonder so many media buyers and sellers are confused and distrustful, causing some to even take a passive stance when it comes to fraud. How can anyone develop a strategy to overcome a threat when the risk hasn’t been properly quantified? Rather than sit back and allow hackers, botnet operators, and pixel stuffers to undercut your online advertising potential, here are three areas to consider when assessing verification vendors and the right way to size up the challenge.
Coverage and Data Distribution
Without good coverage across the web, any estimation as to the amount of fraudulent impressions served will be speculation at best. But what does good coverage mean? No vendor has the luxury of capturing 100 percent of Internet activity, whether with crawlers, pixels, panels, or all three. But a vendor that has visibility into thousands of campaigns that span all categories and industries, performs real-time signal analysis, while also layering historical data for maximum accuracy, can ultimately achieve a true representation of the web. The trick, however, is capturing the correct distribution of data across the multiple sources of display inventory, as each source suffers from different levels of fraud vulnerability.
If a reported level of fraud seems too high or too low, it’s worth questioning whether the measurement is heavily skewed toward a particular type of distribution channel. Full visibility into premium publisher activity but only a keyhole vantage point into long tail sites will paint a picture of low risk. Alternatively, emphasis on exchange and network activity will lean towards high risk. At Integral, we see levels of fraud approximately four times higher for networks and exchanges than for direct publisher buys. With so much activity happening through aggregated and automated channels, fraudulent activity on direct publisher buys barely impacts our overall average.
Sophisticated, Ever-changing Technology
Any inquiries by buyers and sellers into how fraud detection technology actually works will likely garner some general, maybe even vague, responses. Vendors are fighting an army of suspicious and nimble characters, quick to modify their strategies as soon as their covers are blown. The need to keep stealthy combat maneuvers secret is therefore understandable.
Although there may be some common signals that vendors use to detect fraud, fraud detection technologies are far from equal. The ability to not only apply a combination of tested strategies, but also layer in some proprietary secret sauce, will only be effective for a limited period of time. Building and applying fraud detection technology is an exercise in constant innovation. As mentioned, the enemy is nimble, which allows him to move between targets and change his disguise. A reported low level of fraud may actually be the result of subpar technology, or once superb technology that failed to evolve over time. Is the vendor able to identify browsing behavior atypical of humans? Are they looking for abnormal distribution of browser activity? Do they have an arsenal of secret weaponry? Since peeking under the hood is not always an option, use a heavy investment in research and development as a key indicator when appraising the source. A vendor that is able to attract, develop, and retain an elite troupe of data scientists will likely be most adept at making sure the technology stays on the cutting edge.
Placement Level Detection
Perhaps the most overlooked distinction when evaluating your fraud measurement options is whether the technology can detect fraud on an individual placement or impression, as opposed to on a site. This is critical because bot-based fraud happens at a user level. A site experiencing traffic from suspicious sources is typically not void of legitimate traffic. Bot operators may have infiltrated certain pages, generating clusters of fraudulent impressions on specific areas at specific times through specific (infected) computers. Those clusters may represent 2 percent of a site’s inventory, or maybe ten times that; in either case, it’s certainly not 100 percent. Yet, that’s the label that a vendor might slap on a site, using a very binary assessment – all fraud or no fraud – which is not only unfair to the publisher, but can dash an advertiser’s hopes for scale. Higher reports of fraud can be based on overestimations of fraud across a site, rather than across specific impressions, painting a bleaker picture of reality by exaggerating risk, and killing reach and scale.
Consider also the following: it is possible to block fraudulent impressions from ever being served in real time. However, again, the technology must be able to not only detect fraudulent users, but also take action against impressions and not against an entire source of inventory. Lastly, any vendor that argues blocking fraud ultimately informs fraudsters of secret modus operandi is incorrect. It’s very difficult and highly unlikely that fraudsters are able to reverse engineer the reasons why their efforts sometimes fail.
Whether you are a media buyer or seller, ask yourself the following when presented with a level of fraud from a verification vendor:
- Is the data pool representative of the digital ecosystem, capturing data from all inventory sources in the same proportion as their level of transaction?
- How sophisticated is the technology? Has the vendor invested in and retained talent within research and development?
- Does the inability to analyze placement-level data result in overestimations of fraud? Are these overestimations then diminishing scale by generalizing the unsavoriness of an entire site?
When comparing reported fraud levels, assess the source and who may be gaining from skewed measurement. Fraud is an industry-wide epidemic that requires sophisticated, dynamic technology for accurate measurement and risk assessment, and finding the right partner to combat it merits your due diligence.