23 Dec 2014

Inefficiencies in RTB

The inception of RTB into online advertising has brought the industry leaps and bounds beyond how we previously operated. Although many advertisers still do fixed buys on specific properties, they never had the ability to target users rather than the “general audience” of a certain site. Bidding on individual users rather than buying bulk impressions on a site-by-site basis was definitely a game changer.

In real-time-bidding, an auction for a single ad impression is held in approximately 100 milliseconds; meaning there can potentially be 10 exchanges of an impression within the approximately 1 second that it takes for a page to load. The technology behind this is mind-blowing, which you realize even more when you try to explain RTB to friends outside the industry.

Because of the ability to have multiple auctions for an impression, we end up having what I like to call “recycled impressions.” What I mean by that is that even though in reality it is a single ad-impression, because it is traded 5 or more times that leads to it being counted by 5 different entities on a platform like AppNexus. Although AppNexus’ system knows it is a single impression, if we add up the individual impression counts of all the networks on the platform, it will be much higher than the true number of impressions that publishers have plugged into the exchange. Although this does not hurt anyone really, it is a strange concept.

The only issue with having many auctions for a single impression is that it brings middle-men in between the source/publisher and the end buyer/advertiser. Most of the time these middle men are ad-networks that serve the purpose of connecting supply with demand. So Publisher-1 plugs their inventory into Network-A and then the impressions is bought by Network-B on the exchange because it fits their targeting for the demand they have from Network-C. Network-C has a relationship with Network-D to source specific inventory for their agency client or Advertiser-1. So obviously Networks A, B, C, and D all need to be compensated, so there ends up being quite a disparity between the price paid for the impression by Advertiser-1 and the actual amount received by Publisher-1. This concept is very counter-intuitive. Because networks generally work on a rev-share model, the final amount paid by Advertiser-1, trickles down from network to network until Network-B gives their rev-share to Network-A who then gives Publisher-1 their cut. Think about that for a minute. How can we make this process more efficient?
Let’s assume all of these transactions occur in a single exchange. Publisher-1 plugs in 1,000,000 impressions a day to Network-A. Assuming all of the million are sold in the manner I described above, now 4 networks are all showing to have transacted/managed 1,000,000 impressions. So if we take the total counts of networks on a list in AppNexus, it shows that there were 4,000,000 impressions transacted on the exchange. While in reality there were only actually 1,000,000 impressions eventually purchased by an advertiser.

We must acknowledge the role of the networks in this chain, because they spend the time to connect demand to the required supply. If Publisher-1 knew how to find Advertiser-1 on their own, networks wouldn’t exist. But a publisher’s job is to manage their site and attract the audience that advertisers are looking to purchase. Advertisers and media buyers can’t always target as specifically as they would like so much of the transaction is based on relationships and trusting a network to source the supply required for a certain campaign.

RTB is still in it’s youth and we are all looking forward to it’s future development in the environment of online advertising. But at every step along the way, we must take a step back and observe where we see inefficiencies in the process.

Thanks for reading! Please feel free to comment or reach out directly with any questions, concerns or suggestions.