• by Tom Hespos , September 16, 2003

On September 10th, Cory wrote: "Will online ever adopt an upfront season?"

Sure. In some categories, the upfront already exists. But why has a de facto upfront emerged in the automotive category but not for the general market? Cory touched on the reason why in his article last week. It has to do with the dynamic delivery of inventory in our business.

Dynamic delivery is what gives any ol' Run of Site (ROS) ad the ability to become a targeted ad. Whenever an online ad is served, a number of factors are considered by a publisher's ad server that may result in a targeted ad being served. Content area, geography, database profile information, external data, the presence of plug-ins, the price paid for the ad - there are probably hundreds or thousands of factors that can affect which ad gets served when. Sometimes, it helps to think of a site's inventory as comprising a constantly changing number of groups of ad impressions. The number of potential ad impressions in each group is constantly changing, based on the duplication between groups.

For instance, if I buy 80 million ROS impressions on a portal at $1.25 CPM and someone comes along and buys 12 million impressions targeted to people in Florida for a $5 CPM, that buy affects my buy in a number of different ways. First of all, it affects my pool of potential inventory. Geotargeted inventory that commands a higher price will dynamically pre-empt my ROS inventory to some degree. Based on how much inventory the site has and what other campaigns might be running at the time, the Florida buy might introduce a geographic skew to my buy. Depending on the targeting criteria used and a few other factors, including revenue optimizers in use by publishing-side ad servers, any given campaign has great potential to affect another campaign's delivery.

This dynamic element to online advertising inventory delivery might make it difficult to implement an upfront buying structure like we have in broadcast. Where the upfront model has emerged successfully, and of its own accord, is in a category where dynamic inventory allocation has the least effect on things.

In the automotive category, advertisers are typically after high-demand inventory like content sponsorships or very targeted inventory such as keywords. Coincidentally, these inventory types are the least affected by dynamic inventory allocation. After all, if you buy a content sponsorship, you own every ad impression in that content area and anything that another advertiser buys won't affect your buy at all. Similarly, if an advertiser purchases specific keywords, it's not likely that another buy will horn in on any potential keyword inventory and cause the buy to underdeliver.

Can we have an online upfront? Sure, but it will emerge first in the product categories for which inventory allocation is the least complex. Some categories, especially ones that rely on complex targeting to deliver the best-qualified user, will need sophisticated inventory prediction tools to be able to know exactly what can be offered and purchased in the upfront.

One thing's for sure - for an upfront to be successful, buyers and sellers are going to need to watch themselves with regard to overlap between audiences that might cause one advertiser's buy to be pre-empted by another's buy in ways that are not immediately obvious. For instance, my buy against active online investors might have its potential pool of inventory significantly reduced by someone else's purchase of impressions against New Yorkers with HHI $100K+.

Upfront buying can happen - it's happening already. But to be successful, we need to make sure we don't trip over our own targeting capabilities.