• by Yoav Naveh , Op-Ed Contributor, October 21, 2015

Programmatic sale of video advertising today is structured in a way that encourages a highly problematic form of "arbitrage." Here is how it works:

In display advertising, when you win an RTB auction, you commit to paying a price for that impression. That's because display was built on a simple premise: You put in a bid for an ad for $3, and if you win it, then you pay for it. The only notion of “changing your mind” would be through a "passback tag," which allows publishers to receive network or remnant tags to fill inventory when their primary ad network doesn't have an ad to serve that meets the floor CPM. Passback tags in display were not efficient and often led to both a loss of impressions and an increased page load time with 9-12 server redirects for every ad call.

Rather than doing away with display's passback inefficiencies in video, it has become the standard. What most people do not realize is that "winning" a video auction does not necessarily lead to paying for an impression.

In video auctions, it is common practice that a buyer will bid a certain amount, but will then report an "error" in playing the ad, thus not paying for it. There may have been an initial justification for these reported "errors" as part of the verification of inventory quality. However, this practice has led to a risk-free arbitrage whereby the ad network ‘buys’ the impression and then tries to re-auction it with a higher floor in another SSP. (Note: This scenario is certainly possible in display, but in display you at least commit to a CPM.)

For video, without the guaranteed CPM, this is the outcome of the current "passback" setup:

• Ridiculously long load time for remnant tags. By that, I mean an average of 15-20 seconds. (Yes, no milli- hiding in that number…)

• Loss of impressions as users abandon the page (no one will wait that long to show a 15- to-30-second ad) or even just the ad server giving up and skipping the ad placement

• Loss of revenue.

Let me explain that last point, as it is not trivial.  In July, our optimization engine blocked the top 7 bidders (by revenue) using our platform. As a result, overall publisher revenue increased by 35%.

How could that be? Two simple reasons:

1.  While those buyers would bid $10, their error rate was 97%+, meaning their effective “Pay CPM” was $0.30 or less.  A buyer with a mere $6 CPM but under 10% error rate will provide a $5.4 “Pay CPM”.  It became a pretty easy decision to let the optimization engine block the top CPM bid partners once we analyzed the actual pay CPM rather than the bid CPM.  But one could still argue -- why don’t we try the $10 buyer, and if they “passback” (nice way of saying ‘timeout on reselling it’), and then just go to that $6 buyer. Well, this brings us to...

2.  A "buyer" that fails to actually buy the ad takes a very long time (close to a minute) to "not" buy it. Standard timeouts in leading SSPs today are around 10 seconds per ad, each with 5-7 attempts. That is a full minute in most such cases of waiting.

So either the publisher/platform does the responsible thing of timing out the ad call and letting the user watch the content they wanted to watch (essentially) ad-free, or the user leaves the page (frustrated). But even if the user stays on the page reading other parts of the page, and the publisher tries another ad -- the performance is not the same.  Showing an ad in the first few seconds of the page-load performs significantly better as it is more likely to capture the user’s attention. With all of the waiting in the current constellation, by the time the “good” advertiser is actually reached, the impression is "weak and tired" and publishers are pushed to reduce the price and value of their content.

I know that SSPs are struggling to deal with this phenomenon. Publishers need to look at the optimization process and find ways to pre-load ads and better understand the real "Pay CPM" on a page by page, impression by impression level if programmatic is to provide true value to video advertisers.