"Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Augustine Fou, digital strategist and independent ad fraud researcher.
There are many limits to measurement that may lead to entirely different answers depending on when, where and how the measurements were made.
For example, on-page measurement may show a small number of bots, while in-ad measurement of the same campaign may show very high levels of bots.
If measurement vendors don’t disclose to marketers how, what and when they measure and what assumptions went into the measurement, how can the results be trusted?
Technology Limits Lead To False Positives, Negatives And Answers
If a measurement vendor detected mouse movement, page scrolling and clicks, it might mark the visitor as being human, not a bot. But those actions may have been faked by an advanced bot. This would thus be a false negative for nonhuman traffic (NHT) measurement. Or if a fraud detection platform saw a visit coming from a popular data center, such as Amazon Web Services, and declared it a bot, it would be a false positive for NHT if it was really a human accessing the internet through a proxy service.
What if a measurement vendor declared an ad viewable because it was above the fold on the page it was measuring? That page, however, may be in an iframe behind another page, making it impossible for anyone to have viewed the ad. Since viewability measurement technology cannot look outside of the iframe on which it is installed, it cannot detect that the page was entirely hidden. That would be a false positive for viewability.
Or what if an ad was marked nonviewable but the user later scrolled down the page and it became viewable after the measurement was already recorded?
Technology can be tricked or defeated. Good guys’ technology will always be playing catch-up with the bad guys’ technology. When they do catch up, the hackers will always find new workarounds because that’s what they do – they hack.
Measuring In-Network, In-Ad Or Onsite May Yield Entirely Different Answers
Where the measurement is performed has additional limits. For example, in a real-time bid environment – in-network – the measurement vendors might have 10 milliseconds to make a decision on whether to serve an ad. Usually they only have two bits of information to use to make that decision: the domain on which the ad will be served and the user’s cookie or identifier.
The fraud detection service will let the ad serve if it has not seen the domain before or if it has never seen the user before. These services rely on blacklists of domains or users; if either domain or user is not on the blacklist, the default action is to serve the ad. That is because they make money only when the ads are served. In-network measurement is thus limited to catching the most obvious fraud – bots and domains already on a blacklist – and often will not catch anything more advanced.
Another form of measurement is in-ad measurement, where the embed codes ride along with ads as ad tags. This is the most common form of measurement, but it is unfortunately far more limited compared to tracking codes installed directly on web pages by website owners.
The difference is due to basic browser security, which prevents tags in foreign iframes – iframes from other domains – from reading anything on the parent web page. Therefore, those tags cannot determine where the ad iframe is on the page, which is necessary for viewability measurements – is the ad above or below the fold?
The tags also cannot read any content on the parent frame, which is needed for brand safety checks. So how does a vendor measure brand safety or keyword targeting when the embed code is in the foreign iframe of the ad? Marketers should ask.
Challenge Everything, Even If You’re Being Told What You Want To Hear
Even if a vendor gives marketers the results they want to see, they should always ask for more details so they can verify their accuracy. Many already do this when reading Yelp or Amazon reviews, so why not also apply the same scrutiny to results from measurement vendors on NHT, invalid traffic, viewability and everything else in between?
Digital media buyers named data transparency as their No. 1 criteria when evaluating ad tech in a recent study. They should be able to verify how the measurement was made and why a particular visit was bot, human or viewable. Most vendors do not provide this level of transparency and detail. They won’t do so until marketers insist.
Marketers should ask more questions, harder questions. Ask for line-item details because fraud hides easily in averages. Ask for the number of data points used in the measurement and if the answers were extrapolated from a tiny sample. Ask if any assumptions or approximations were used in the measurement. When when the measurements were performed and where – in-network, in-ad or on-page. Remember that the most common place of measurement – in-ad – is also the most limiting.
Constantly challenging measurement results will make marketing campaigns more efficient and help advertisers understand how accurate and trustworthy their measurements are.