The State of Ad Serving

I had the good fortune to sit on a panel a few weeks back at the Chicago Interactive Marketing Association’s (CIMA’s) May meeting. CIMA recruited product people from the top ad-serving companies to cover ad-serving issues. Today, I’ll discuss a few issues raised, specifically discrepancies and inventory control. I’ll also cover a topic that didn’t get much play: rich media.

First, let’s get our definitions clear (as always). I’ll refer to two types of ad servers: third-party ad servers (3PASs) and site-side ad servers (SSASs). 3PASs are used by advertisers and agencies to serve ads across multiple publishers. 3PASs allow them to consolidate reporting in one interface, capture post-impression and post-click activity (conversions), and streamline the ad-trafficking workflow significantly. Examples of 3PASs are DoubleClick’s DART for Advertisers, Bluestreak, Atlas DMT, and Mediaplex.

SSASs deliver ads to publishers’ Web sites and deal with inventory control, scheduling, and delivery. Examples of SSASs are DoubleClick’s DART for Publishers, 24/7 Real Media’s Open AdStream, Fastclick, and ZEDO.


This was the single hottest 3PAS issue a few years ago. Essentially, all 3PASs see some kind of discrepancy when comparing their impression numbers to the publishers’. It’s the nature of the Internet. The publisher counts an impression when the ad tag is served to the page. The 3PAS counts the impression when the ad is sent to the user’s browser. This time lag — sometimes only a few milliseconds, sometimes longer, depending on many factors — is unavoidable. It leads to some sort of discrepancy in almost every case.

An acceptable discrepancy is about 10 to 15 percent. All of us claim average discrepancies in the single digits; I know we average about 3 to 4 percent. We’ve all seen legitimate cases around 20 percent, such as with an inherent discrepancy between two systems. But that’s rare. At a customer’s request, we’d investigate a reported discrepancy above 10 percent.

The discrepancy rate has been dropping for a long time. We’re at the best efficiency we can expect, outlying cases aside (for the moment.)

Why has the discrepancy rate dropped? A few reasons:

  • Increased broadband usage. The primary cause of legitimate discrepancies is the time lag from the publisher count to the third-party count. The faster the connection, the shorter the time lag between servers, and the less likely the user will cancel the action or close the browser.

  • More standard ad servers and counting methods. A few years ago, the Interactive Advertising Bureau (IAB) and American Association of Advertising Agencies (AAAA) published a counting standard. It’s taken a long time for many smaller publishers to adopt it, but most have now. That means we’re (mostly) all counting the best way we can. Certainly some publishers still use page impressions rather than ad impressions to count ads. But most do it properly, which alleviates a huge problem.
  • Industry maturity. We’ve been serving ads for a while now, and there are fewer user errors on both sides when implementing ads.

Inventory Control

A SSAS inventory control system has many tasks. One of the most important is to forecast available inventory for the publisher’s ad sales team. If the system isn’t doing a good job, the site could sell more ads than it can actually run. Or, it might think a site section is sold out, then end up with lots of vacant space.

During the dot-com bubble, poor inventory control was one of the biggest SSAS complaints. Inventory was almost always “sold out,” leading to huge problems with makegoods to advertisers.

Say an advertiser buys 1 million impressions to run on a site during May. The site delivers only 500,000. The publisher then has to fork over half a million impressions in June for free. This creates a circular problem, because June’s inventory is tightened up. Besides, the advertiser planned on those impressions in May to hit sales goals. If the publisher underserved ads, the advertiser may not get the needed value from those impressions.

This problem eventually dried up, post-bubble, when publishers generally weren’t selling out their inventory. Rather than fix the problem, the industry let it disappear. Now we’re successful again, and inventory is becoming scarce. We’ll be right back where we were in 1999 and 2000, as the problem was never fixed. It should be interesting to see what happens in the next few years.

More interesting will be what increased demand and decreased supply does to media costs. We’ll be holding our breath on this one.

Rich Media

All 3PASs now internally support some rich media functionality, such as Bluestreak’s Advanced Flash Tracking and Rich Media templates, DoubleClick’s DART Motif, and Atlas DMT’s recently acquired layer-based ad provider Ad4Ever.

In addition, these servers support to some extent existing rich media technology providers, such as Eyeblaster, Viewpoint, Unicast, and PointRoll.

Rich media technology providers also license their technology to publishers and advertisers/agencies directly, though typically publishers pass the cost through the media buy. In some cases, advertisers buy ads from the rich media companies directly and just place them with the site.

Ad serving has been interesting lately, but where rich media will ultimately reside remains to be seen. Will rich media providers continue to add CPM onto the media buy, or will they move to a licensing model?

Related reading