Programmatic’s Viewability Crisis | Unpacked

“When an executive at a leading exchange claimed 100% viewability…that’s when the intellectual brawl began,” said C3 Metrics advertising attribution measurement CEO Mark Hughes.

“It was a private dinner event held by the research division of a Wall Street bank. The table stretched 45 feet with equity analysts mixed with digital advertising insiders. The discussion was meant to educate the analysts, providing them with expertise that’s hard to come by.”

As the debate got deeply technical and voices escalated…silence engulfed the room as the Wall Street analysts watched the Main Street slug-fest.

The programmatic exchange exec wouldn’t let go of his claim: “100% viewability.”

His stubbornness highlighted the complexity of viewability, even for an expert.

But today’s Digiday article by Ross Benes on how Ranker.com doubled its viewability rate in 15 months, is one of the best articles beginning to unpack the many components of viewability.

Viewability = Publisher + Exchange + Consumer:

Publishers

This is what Ranker.com did to change and double its viewability (it took over a year):

  1. How ads are rendered
    • Are ads rendered via Slideshow (especially terrible for mobile web viewability).
  2. Page length
    • Long pages = terrible for viewability because ads are below fold. Short pages with pagination (1, 2, 3) are better for viewability.
  3. Page load speed
    • If a page doesn’t fully load, the ad won’t load either. Page load speed is different in desktop and mobile (even for the same page). Slow speed also gives the consumer time to click to another page before the page fully loads, and thus the ad will not be viewable if the consumer clicks away.
  4. “Lazy-Loading” or “Smart Loading” Ad Technology
    • Publishers using “Lazy-Loading” ad technology enable ads to be called with an async HTML5 attribute which is simply a way to conserve browser resources.
  5. Using Slow Exchanges
    • All exchanges are not created equal. Different ones have different load times, and also load more info. Imagine running the 50 yard dash, then running it with two jugs of water.

Exchanges

The IAB definition of an impression is simply an auction win; the display unit has not yet been delivered to the user’s page/device.

  1. Impression ≠ Ad Served
    • If an advertiser, through the exchanges, bids and wins on the right to deliver a programmatic display impression, the impression is defined as the auction win. The ad has not even been served yet, it just means that it’s been bought.
  2. Display “Views” are Not Views (just cookie drops)
    • Although an impression is simply an auction win, the ad then gets called from the server side and then has to load in the consumer’s browser viewport. Once it loads, it simply drops a cookie, and cookies have nothing to do with viewability.
  3. Age + Architecture = Different Speed
    • Different exchanges, like cars, have different technology, reliability, and speed. Ranker.com got rid of the slow and heavy exchanges it was using, and only allowed the fast technology of modern exchanges. If you drive a Pinto across the country versus a Corvette, you get there at different times. 20% of the time, display ads never load in the consumer’s browser because the Pinto never makes it across.

Consumers

Consumers are the third variable in the Venn diagram affecting viewability (and neither publishers nor Exchanges can control a consumer).

  1. Consumer’s Connection Speed
    • If a consumer is on WiFi, or a slow 3G: it makes page load speed very different. This is tangential to the load speed of a Publisher’s page (as well as an Exchange).
  2. Consumer Scrolling/Clicking Behavior
    • If a consumer scrolls down a page faster than the ad can load, or clicks away from the home page to another page on that same site, user behavior can be faster than ad loading. Neither Publishers nor Exchanges can change how consumers scroll or click.
  3. Browser Viewport Sizing + Applications Covering the Browser Viewport
    • Browser Viewports (not necessarily the device screen) can be stretched, squeezed, or shrunk. Browser viewports can also have things on top of them covering ads, for example: a second browser viewport, or an application like Outlook, Excel, and Powerpoint. Neither Publishers nor Exchanges can control the environment in which a consumer chooses to consume media.
  4. Device Type
    • Which device a consumer uses affects both their scrolling and navigation behavior.  On desktop/laptop the speed of clicking links with mice is dramatically different than mobile which is navigated more with swiping.  Mobile versus desktop/laptop usually manifests different rendering because screen dimensions are also different.  But neither Publisher nor Exchange can control consumer’s behavior to use Ad Blockers.

The basics of measuring viewability:

  • Has to be done in the cross-domain iFrame
  • Has to be done in four browsers for desktop
  • Has to be done in four browsers for mobile web
  • Has to be done using the IAB guidelines for display and video

In C3 Metrics case, all of this matters (and is done in milliseconds). To boot, C3 Metrics has open-sourced portions of its viewability code to the IAB SafeFrame initiative.

But back to the geek brawl in front of the Wall Street analysts: add this up and there are 12 components within three legs of the viewability stool.

The golden rule of understanding viewability (which everyone forgets) is that impressions are clocked at the server/Exchange side, but viewability is clocked at the consumer side (in the consumer’s browser viewport).  The loss factor from one side to the other is at least 10-30% because of Exchange factors (the ad never loads in the consumer’s browser).

Then take any combination of the 12 components in this Venn diagram…and nobody gets 100% viewability. Nobody. Claim 100% viewability, and you will get schooled.

But it’s an intricate problem to solve: C3 Metrics began working on viewability in 2009 because it’s mathematically crucial to user-level attribution. Here’s the math:

  • 95% of all data collected for an advertiser are programmatic “display views” but if 68% of those views are not viewable, then 66% of your data is wrong before you even begin to attribute/Bayesian model it.
    • [95% x 68%] = 66%

Not many CFO’s ask for data to compile their earnings, and say “but make sure it’s 66% wrong before you give it to me.”

If more Publishers like Ranker.com went through the effort to understand, unpack, and make changes to their sites, this could be the end of the programmatic viewability crisis.

Pack your bags.