The most common unit of measurement in digital advertising is the impression. And, it’s the most common way that digital advertising is bought and sold today. It’s so commonplace that most people take for granted that they know what it means. Well, most people are wrong.
Most of us think that an impression means that an ad was shown to a consumer. They may not look at it, but they have the chance to see it, right?
Unfortunately, that’s wrong. An impression means one and only one thing. It means that an ad server was called. The official IAB definition reads:
“‘Impression’ is a measurement of responses from a Web server to a page request from the user browser”
Technically, here’s what happens: whatever ad server you use provides a “pixel”—a tiny, usually invisible image—that lives on each publisher page. When this image loads, the ad server is notified, and you get an impression. But an impression does not equal an ad opportunity. Three things get in the way:
The ad is broken. No foul play here, just means that the wrong ad loaded, or something technical went wrong and the ad didn’t appear. Most ad networks will report, and try and manage, these numbers but it happens about 15% of the time.
Hackers write bots, automated computer programs, to crawl the web for various reasons, like posting spam and creating phony accounts. These bots count as site views, falsely driving up the impression count of your ad. An ad server is called even though no human is on the site. Multiple studies show that up to 60% of all traffic on the web is bots.
There is lots of money to be made in purposefully creating fraudulent impressions to steal ad dollars. Recently, a fake web trafficker came clean on the gory details. People will hide ads behind other ads, spoof their domain to trick ad networks into serving higher-paying ads on their site, and purposefully send bots to a site to drive up impressions. Wenda Millard, president of MediaLink, claims that 25% of the entire online ad market is fraudulent. The end result is lots of traffic and lots of inaccurate measurement.
Most ad networks do want to control this, but with hundreds of thousands of sites involved, it’s impossible to police everyone. What you’re left with are some staggering numbers.
We start with the notion that only 15% of impressions ever have the possibility to be seen by a real person. Then, factor in that 54% of ads are not viewable (and we already discussed how flawed that metric is), and you’re left with only 8% of impressions that have the opportunity to be seen by a real person. Let me clarify: that does not mean that 8% of impressions are seen. That means only 8% have the chance to be seen. That’s an unbelievable amount of waste in an industry where metrics are a major selling point.
This begs the question, why do we pay so much attention to a metric that isn’t meaningful? In part, it’s because impressions are a very easy metric to measure, even if the information they provide ultimately isn't that useful. Furthermore, we've been measuring impressions for years, and that inertia is hard to overcome. But as a smart guy once told us: “Not everything that counts can be counted, and not everything that can be counted counts” —Albert Einstein.
Let’s start trying to count what actually matters. What we really want to know is: was a brand message actually seen by a real live person? Measuring impressions, however easy it may be to do, doesn't answer this question. Measuring engagement does. As a first step, we should only count impressions that we know aren't fraudulent, and that we know aren't bot traffic. But to do even better, we should aim beyond impressions, to metrics that can tell us not only that a person saw an ad, but that they actually interacted with it.
There's no question that this will be more difficult than the old way of simply counting impressions and calling it a day. But if we push towards measuring what actually matters, we'll make everyone—advertisers, publishers, and consumers, too—happier.
Great analysis of the most overused but misunderstood term in advertising.
Regardless of the many different definitions of an 'impression', the fact is that it is a flawed and outdated metric for measuring ad performance. Measuring a campaign success on 'impressions' can only give insight on volumes, and provides no data for quality of placement or consumer engagement. We should be delving deeper into the anatomy of a page that surrounds the ad, using relevancy and contextuality as a measure of quality of placement and user engagement to measure campaign success.
Good job, Reid. I liken an impression to a digital tree falling in the forest when no one is around to hear it. But in all fairness, the ad sellers already know this. That's why impressions can be had for sub-pennies.
It looks like you have a lot of overlap here. You include bot traffic and fraudulent traffic that includes bots, so you cannot simply add those percentages together.
IAB standards require members to exclude bot traffic from reporting and billing numbers based on IP addresses, user agents or other criteria so you shouldn't suggest that 60% of billed impressions are being served to bots. Most bot traffic is being identified and removed from reported numbers. Fraudulent bots are more difficult because they try to appear legitimate, but those can also be detected through analysis and pattern recognition.
Finally, viewable impression measurement also takes into account unviewed impressions due to bots or hidden ads so those numbers cannot be simply multiplied together.