The Ironic Sophistication Of Viewability

Why do marketers buy display advertising? Why do they deal with such a complex programmatic ecosystem? Why do marketers do any marketing at all? It seems reasonable to conclude that no matter which tactic marketers use, with so many choices at their fingertips, the answer should boil down to one simple objective: to connect a product or brand with its audience.

However, when programmatic marketers attempt to achieve this mission, poor impression quality will stand in the way of making a real connection.



Working toward a market-priced solution

Let's start by laying out a few reasonable assumptions:

  • Forrester's recent DSP Wave highlights the sophistication of DSP algorithms.

  • DPSs are capable of incorporating many different data types into their algorithmic recipe.

  • DPSs typically claim algorithmic sophistication in their sales pitch and marketing materials.

  • Some DSPs are better than others (e.g., home-grown tech vs. those built on top of second-party generic platforms).

If you agree so far, then you will hopefully buy into a few additional assumptions:

  • Not solving the problem of lesser impression quality could cause marketers to seek substitute methods of connecting their product and audience.

  • Over the short term, many parties, from buy-side to sell-side, will suffer some level of financial loss as impression quality solutions work their way into the market (e.g., many  players today benefit financially from poor impression quality).

  • Over the long term, the exponential value creation of creating a clean marketplace will outweigh the short-term cost.

Solving viewability in the bid

Perhaps the most significant aspect of controlling impression quality is how (relatively easily) it can be solved with the aforementioned DSP sophistication. Let's start with a simple ratio called impression quality. If quality is based on the "opportunity" for an ad to be viewed, then a simple quantitative measure of success can be expressed as:

impression quality = opportunity to be viewed ÷ served impressions

In order to totally maximize this success metric, buyers should only bid on impressions that meet the opportunity criteria. In other words, avoid bidding on impressions you would not ad-serve in the first place.

Given all the DSP sophistication, it seem there should be at least a few DSPs that incorporate and promote deterministic or probabilistic factors into their algorithm in order to price user-impression quality into the bid.

Imagining how this might work

Imagine a typical bidding situation as it is today. A DSP sees a user-impression combo and calculates a user valuation of $5 on a floor price of $2.25. The DSP bids $5 and wins the impression.

After serving the ad, the post-hoc data comesin, telling the marketer that the impression was either “below the fold,” served to a machine not operated by a human, blocked by ad-blocking software, and/or served to a background browser. Had this information been predicted beforehand by the DSP, the valuation would have been calculated at $0 -- no bid would have been made and no impression served.

Now, imagine a DSP sophisticated enough to include either a deterministic or probabilistic "quality factor" before making the bid by ingesting new forms of first-, second-, and/or third-party data.

The probabilistic pre-bid checklist and answers might look look something like this:

  • probability the impression is below the fold = 80%

  • probability the impression is served to non-human = 50%

  • probability the user uses ad-blocking software = 10%

  • probability the impression originates in a background browser = 20%

If we weight all factors equally, the average pre-bid quality factor is 40%. Therefore the bid price is now 40% x $5 = $2. No bid is made, no ad is served, no money changes hands. Such a pricing mechanism would effectively mean classifying impressions as non-biddable or biddable.

Now let's see how the more powerful deterministic pre-bid checklist might work, where deterministic probability is expressed as a binary function equal to 1 or 0, where 1 = biddable and 0 = non-biddable.

  • probability impression is above the fold = 1

  • probability impression is served to a human = 1

  • probability the user does not use ad-blocking software = 0

  • probability the impression does not originate in a background browser = 1

In this all-or-nothing setup, all the questions must be answered with a value equal to 1. If not, the bid equals $0. No bid is made, no ad is served and no money changes hands if the impression is deemed to be non-biddable.

Impact of correctly priced markets

By using this pricing valuation method, the impression quality problem gets solved in a very democratic way for both buyers and sellers. Publishers and their SSPs would have a market-driven incentive to control inventory quality, plus the extra incentive to provide more data for DSPs to calculate accurate pricing. The more advanced publishers would find ways to put similar algorithms to work and adjust CPMs accordingly.

Moreover, the MRC, IAB, ANA, 4A, etc, plus all the other stakeholders involved, could repurpose their energies toward bigger problems such as true market pricing indices and education initiatives that grow new talent. In the end, both advertisers and publishers could spend more time finding ways to connect product with audience -- and hopefully use programmatic display ads to get the job done.

Request for 'Sell-Side' Contributors

MediaPost is inviting candidates to submit column ideas for a new publication called the Sell-Side Insider, which will focus on business issues related to a sell-side perspective in the digital media marketplace. Interested parties can email
1 comment about "The Ironic Sophistication Of Viewability".
Check to receive email when comments are posted.
  1. Victor Ortiz from InMediaRes, September 10, 2015 at 6:37 p.m.

    Tom, isn't the problem here that the deterministic approach is still based on probalistic measures?
    Is there a way to truly determine viewability deterministically?

Next story loading loading..