The "Dynamic Allocation" Feature Within DFP / DART Is Not Yield Optimization

Google wrote a whitepaper a couple months ago called Profiting from Non-Guaranteed
Advertising: The Value of Dynamic Allocation & Auction Pricing for Online Publishers, which describes the benefits of the "Dynamic Allocation" feature built within DART/DFP. Dynamic Allocation is a mechanism for (a) opening a publisher's non-guaranteed ad inventory to the DoubleClick Ad Exchange (Ad Ex), and (b) giving Ad Ex the right of first refusal on that inventory.

I understand Dynamic Allocation works like this. When publishers book a non-guaranteed campaign into DFP (meaning they have not guaranteed a fixed number of impressions), they can check the Dynamic Allocation box. Then, any time that campaign is selected by DFP for display to an end-user, DFP will first check with Ad Ex to see if it can beat the CPM value entered for that campaign -- if Ad Ex can beat it, then the Ad Ex campaign will trump the campaign that the publisher booked.

The problem is that DFP uses a random selection algorithm. DFP, like all other ad servers, is designed to deliver guaranteed campaigns (meaning a fixed number of impressions to be delivered within a set timeframe, at a set CPM rate), not non-guaranteed campaigns. DFP's decision-making process with regard to which non-guaranteed campaign to serve, is random -- not based on yield (revenue to the publisher).

"Ad servers like DFP have some variability in how they meet impression delivery goals, rotating other ads sold by the direct sales force on a non-guaranteed basis into a particular ad slot in order to maintain a steady rate of delivery for the guaranteed ads. Instead of randomly rotating other ads into an ad slot, DoubleClick Ad Exchange uses Dynamic Allocation, rotating in higher-paying ads from ad networks and other third-party media buyers when the net CPM they provide to the publisher is higher than what has been booked directly into the ad server."

Why is this important? Because without yield optimization in the ad server itself, Ad Ex doesn't have to offer the most competitive price every time in order to win the impression. Which means that Dynamic Allocation often costs the publisher revenue instead of making it more.

I'll back this up with an example. Let's say Publisher.com has 4 non-guaranteed campaigns booked within their DFP ad server in the same tier, and the expected CPM rates for those campaigns are:

Campaign	Expected CPM
Campaign 1	$.50
Campaign 2	$.60
Campaign 3	$.85
Campaign 4	$1.20

In its effort to deliver guaranteed campaigns on an even basis, DFP will draw randomly from these 4 non-guaranteed campaigns. 25% of the time it will select Campaign 1, at which point Ad Ex only needs a $.51 bid to win the impression, even though Campaign 2, 3 or 4 would have made Publisher.com more money. If Publisher.com's ad server optimized for yield, then that would mean the highest paying campaign would serve every time. (For simplicity sake, I'm ignoring other campaign attributes that may limit delivery -- such as geo-targeting, frequency caps, etc.)

Google's whitepaper is clever and convincing, but don't be fooled. While Ad Ex (or any auction for that matter) allows the highest bidder to win, the fact that DFP isn't selecting the highest-paying campaign every time means that publishers utilizing Dynamic Allocation are leaving money on the table. They need true yield optimization instead.

(The above observations/opinions are my own, and do not necessarily reflect  those of my employer.)

Comments

[Jon]

Jordan,

There is a factual error in this post surrounding non-guaranteed ad selection in DFP. Our platform actually does allow publishers to yield optimize by setting these ads to compete on price. Therefore, the Ad Exchange is always able to compete with the best available non-guaranteed campaign in real-time to maximize revenue for publishers. In your scenario, Campaign #4 would always be called first to compete with the Ad Exchange via Dynamic Allocation.

Sincerely
- The DFP team

[Jordan]

Jon, can you explain exactly how publishers can set up yield optimization within DART? The publishers I speak to are only aware of manually setting campaigns into one of a few priority levels -- they're not aware of automatic eCPM/yield-based delivery.