"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 Ted McConnell, an independent consultant in the digital marketing space.
We only know the full truth about how viewable an ad is after the page on which it appears finishes running. Before then, the viewer might scroll to a place where an impression qualifies as in-view. Knowing in advance if an ad will be in-view is impossible – almost.
How can we bid correctly for an impression without prior knowledge that it will be in view?
AppNexus announced last week that it would provide historical placement-level viewability in bid requests, which, considering its scale, has the potential to change the digital advertising landscape.
For the most part, uncertainty now rules because ad buyers don’t have the ability to tell whether an impression will be in view. So those in the industry use a variety of common countermeasures, such as alternate metrics and bidding strategies.
These are all better than the bad old days, but providing historical placement-level viewability in bid requests would include the in-view average for the exact ad that buyers are about to bid on. This is accurate to within a few percentage points, because page design and user behavior are consistent.
If a specific placement is in-view 90% of the time historically, it is very likely to be in view 90% of the time in the future.
Patches For A Disjointed System
Previously, marketers used other metrics and strategies to discern viewability pre-bid. One countermeasure is the vCPM, where the advertiser only pays for viewable impressions. Providers can keep delivering until an impression goal for in-view impressions is met. Many not-in-view impressions are bought, but the end user does not pay for those explicitly. Usually, the CPM paid by the advertiser increases to compensate. This has become so standard that Mediaocean and Moat have a deal whereby the in-view portion of a campaign becomes the basis for billing.
Another tool used by marketers is to bid for impressions that are supposed to appear “above the fold.” An above-the-fold indication is provided in some bid requests. Above-the-fold bidding not only fails to predict viewability; it devalues a huge swath of perfectly good in-view impressions because they are not above the fold. People scroll a lot. It also favors publishers who compromise the viewing experience by cramming all ads above the fold or use tricks like “click here to see the next paragraph,” which makes for a bad user experience.
Marketers can also bid for page-average viewability, or the average of all ads on a page. The problem is that averages mean little in a skewed distribution, and the distributions at page level are extremely skewed. For example, in a typical page, one impression might be in-view 90% of the time consistently, and another might be in-view 10% of the time consistently. This method is a little better than a guess, but not by much.
Finally, it is possible to probabilistically predict viewability by modeling variables associated with a page or impression. Screen size, for example, is a variable that predicts viewability. This works to varying degrees, depending on the skill of the modeler and available data.
To date, the industry hasn’t done historical placement-level viewability in bid requests because it’s hard and expensive. It requires a consistent name reference per ad frame, page geometry, in-view data and so on. That means storing real-time data for every ad on the web, while containing historical averages of viewability for each one. The probability is then placed in the bid request, calibrated by modeling data. Even in a big data era, that’s huge.
While others may follow in the footsteps of AppNexus, the ubiquity and reach of AppNexus, combined with pre-bid placement-level viewability, will change the landscape. AppNexus will also apply predictive modeling so that screen size and other factors further improve on the historical in-view data. I’m told that initial testing at AppNexus suggests the number in the bid request will be 95% accurate.
The stabilizing effect in the marketplace will be significant because no one will bid for impressions with very low viewability. Attribution models will get better, and optimization will become way less noisy. Scam impressions will show 0% viewability.
Now we will be able to abandon the defeatist idea that buying 30% not-in-view impressions is somehow par. RTB will take another step toward the vision.
Advertisers will still have to measure viewability independently. It’s still real money, and in all media, accountability is required. Advertisers still need to know if they got what they paid for and where they got what they paid for, using third-party measures.
With this foundation, viewability will become table stakes, and other indications of quality will become differentiating and biddable. That includes: attention, as indicated by browser or app signals; clutter, as indicated by ad or pixel density; human-ness, as indicated by NHT signals; and context adjacency, as revealed by smart semantic analysis.
The problem with RTB has always been that bidders could not know exactly what they were bidding on. The addition of a key quality element accurately represented to bidders vastly improves relevant knowledge about an impression.
Follow AdExchanger (@adexchanger) on Twitter.