MediaMath COO Ari Buchalter, PhD in iMedia
How to apply real-time benefits to premium ads
Over the past few years, buyers of online display advertising have been witness to an ever-widening gap between the way "premium" and "remnant" advertising is purchased. On the premium side, the process is characterized by high levels of interaction with publisher sales forces, PowerPoint presentations, pricing negotiations, emails, faxes, and Excel reports. But on the remnant side, a once-similar process with respect to ad networks has given way to the emergence of ad exchanges, real-time bidding (RTB), and demand-side platforms (DSPs). These technologies have put greater media and data access, more precise targeting, algorithmic optimization, turnkey automation, and better analytics and insights directly in the hands of buyers, empowering them not only with knowledge of precisely what media and audiences to buy and at what price, but also with the tools to make it happen. Consequently, we've seen the emergence of a new breed of media buyer (typified by agency "trading desks" like Cadreon, Accuen, and Varick Media Management) focused on systematically mining improvements in both efficiency and effectiveness on behalf of advertisers, with many of the largest display advertisers building out similar capabilities internally. It's nothing short of a revolution in the way online media is bought and managed, but it hasn't applied to premium media, until now.
Of course, the premium vs. remnant distinction is an outgrowth of the primary vs. secondary market structure — essentially replicated from the world of offline media. Publishers sell as much as they can up front for higher cost-per-impression (CPM) prices, along with benefits such as content integration, guaranteed delivery, and advance knowledge of where each ad will run. The unsold inventory goes to an aggregator (historically an ad network), which monetizes it at much lower CPMs and, typically, without these benefits to the buyer, but protects publishers from channel conflict by not disclosing specific publishers and prices.
In that sense, the publisher classification of inventory as remnant is somewhat arbitrary since it is, by definition, whatever could not be sold on a premium basis. However, an advertiser's definition of premium and remnant might be very different. For example, an advertiser might say that any impression that targeted its ideal customer would be premium, even if it wasn't on the publisher's homepage or integrated with the publisher's content. Or that any impression that was more than a five-times average likelihood to generate a conversion was premium, even if it couldn't be known in advance on exactly which page it would run. And in fact, much of the value created by the DSP revolution has come from exactly that — buyers using data and advanced analytics to understand which impressions are truly valuable to them and why.
The notion that premium versus remnant is a relative concept is further supported by the data: CPMs for remnant impressions on ad exchanges are higher, not lower, than corresponding CPMs for the same inventory on ad networks. Why? Because DSPs offer buyers a full view of the media, and help the user to understand exactly what works and what doesn't. They can much more accurately separate the wheat from the chaff, and in many cases one advertiser's chaff is actually another advertiser's wheat.
Those DSPs employing sophisticated bid optimization algorithms to value impressions (which isn't all of them, so I encourage buyers to ask their DSPs to see the actual output of those algorithms) are not looking to find cheap impressions that are worth very little (that would actually be a losing strategy); they are looking to find effective impressions that are worth a lot. That's why in many cases the CPMs bid for remnant RTB impressions actually exceed what publishers get from their premium pricing.
As a simple example, consider two otherwise identical ad impressions on the same publisher, but with one key difference: The first impression is for a user that the advertiser knows has visited its site and checked out the product but not yet purchased, while the second impression is for a random user with no such history. That information alone could warrant, say, a $20 CPM bid for the first impression versus a 50-cent CPM bid for the second. Now imagine every advertiser looking at the world through that lens, and imagine they are looking not at one, but hundreds or thousands of data points on every ad impression. That's a very different lens for understanding value than the one publishers look through when they see the same impressions.