Daisy chaining started because early ad servers could not manage a publisher's non-premium (aka unsold) inventory effectively. In the absence of a better solution, publishers started passing along ad calls to multiple partners: For example:
1. Publisher uses a rule-based ad server to deliver its advertiser direct campaigns to its users, but 40% of its inventory goes unsold. The server doesn't manage remnant inventory well, so....
2. Publisher strikes a deal with Network (NW) where NW gives them a flat CPM (say $3) for 300×250s. The deal is struck based on NW reporting and publisher starts sending NW ad calls.
3. Based on the $3 CPM, say NW can only take about 60% of the ads (it's a number game - NW has x salespeople bringing in y campaigns with z budgets - sometimes there are mismatches). To manage this unsold inventory, NW gives the publisher the option of showing a PSA (no revenue), sending an impression back to the publisher or redirecting it to another network. Publisher doesn't want to be charged twice for a remnant ad call, so it chooses to pass the impression to another network.
Other networks and publishers followed and the daisy chain was born. It became an accepted practice. But if you look deeper, you will find some big issues. Here's an example roughly based on a real publisher.
1. Publisher sends 1,000,000 ad calls to Network 1
2. Network 1 receives/counts 972,405 of the ad calls
3. Network 1 delivers 663,436 paying ads
4. Network 1 redirects 308,969 ad calls to Network 2
5. Network 2 receives/counts 268,495 of the ad calls
6. Network 2 delivers 123,549 paying ads
7. Network 2 redirects 144,946 ad calls to Network 3
8. Network 3 receives 121,202 of the ad calls
9. Network 3 delivers 72,149 paying ads
10. Network 3 redirects 49,053 to Google Adsense, which the publisher sticks on the end of the daisy chain because they say they fill 100% of inventory with the AdSense blind rev share deal.
Now the math: through the redirects between parties, there are discrepancies. Of the original 1 million ad calls sent by the publisher, 91,813 ad calls are "lost" between ad servers. Roughly 10% of the ad calls don't generate any money. Not good! Like passing sand between you and a friend when you shake hands, you're going to lose some ad calls when different technologies try to talk to each other. Another reason is latency. Bandwidth issues with any of the networks slow ad calls down and many of them drop.
Latency also decreases user interaction. A marketer has about three seconds to make an impact on a user. If the ad call is redirected multiple times and an ad is delayed from fully serving, the marketer loses an opportunity. Since most of the advertisers that buy from networks are ROI-focused, latency affects the advertiser's ability to pay higher rates. Ever wonder why there are so many distracting ads? Many are a byproduct of latency. They are the marketer's online "Hail Mary" to get the user to respond. Aside from discrepancy and latency issues, there's a bigger problem.
Until recently, networks did not coordinate for the benefit of the publisher. Every online network acts as an economic market maker. It has advertisers with demand and publishers with supply. If a publisher sends a network an ad call and doesn't let other networks compete for the impression -- networks that may be willing to pay more -- the publisher has no idea if it's maximizing revenue on that impression. Sure, publishers determine position in their daisy chain by the aggregate eCPM a network will pay, but not on an ad call by ad call basis.
In my example above, there are cases where Network 3 could pay more than Network 1 for an ad impression (maybe they have an amazing travel deal for Hawaii for anyone in California, and are willing to pay much more for California inventory), but they don't get to "see" the ad call. Network 1 pays the publisher a $1.75 eCPM for an impression that Network 3 would have bought for a $3.50 eCPM. This happens a lot.
But when an open auction is used and where multiple networks compete on an ad call by call basis for publishers' inventory, publishers generate higher yield.
In January, when we looked at the traffic across our exchange of roughly 75 networks and 18,000 publishers and advertisers, traffic was roughly 130 billion impressions with 80 billion unduplicated impressions (impressions served from advertiser to publisher directly through the publisher's network). The remaining 50 billion impressions served from one network's advertiser to another network's publisher.
This is big -- when networks had an equal opportunity to show their own advertiser or show another network's advertiser, 40% of ad calls were diverted to an advertiser from another network. Since most networks work with publishers on a rev share basis, it doesn't matter that another network's advertiser was shown. What matters is that the publisher maximized yield.
With all the issues that daisy chains create, and with the understanding that even the largest networks are often not capable of maximizing the value of every ad call, the question is why publishers use them.
Publishers can and should continue to work with networks that bring relationships and value to the table, but they must begin to reassess their non-premium strategy. Using daisy chains satisfied the initial hunger for monetizing inventory, but it created some tough issues and keeps publishers from developing new buyer relationships, and from maximizing yield on each ad call.