The war on header bidding
This is the latest post of my new Marketecture newsletter: The Monopoly Report. Each week we will cover everything relating to Google’s ad tech antitrust trial. Subscribe for free here.
In the inaugural issue of this newsletter we covered the acquisition of DoubleClick’s DFP ad server, and how that product, in conjunction with the new-at-the-time ad exchange, gave Google a unique competitive advantage called “dynamic allocation.” The Tl;dr is that Google could bid on publisher inventory in parallel with other demand sources, including the publisher’s direct salesforce, whereas other exchanges and ad networks could only bid in serial through what is various referred to a “waterfall”, “daisy chain”, or in the mobile world “mediation.”
This advantage came under attack by clever publishers and vendors who invented and deployed a solution called header bidding. The story might end there, but then there might not be a DOJ antitrust case. This is the story of header bidding and Google’s efforts to thwart it.
The emergence of header bidding
If you want to start a fight on ad tech twitter, just ask who invented header bidding. I’ll spare that discussion, but if you are bored, Digiday did an oral history of header bidding .
Let’s dumb this down as much as humanly possible.
Before header bidding:
After header bidding:
This new way of running auctions put third party exchanges on a roughly equal playing field with Google, and also almost uniformly increased publisher revenue since demand from third parties was accretive to Google demand, which previously had an edge.
Pre-history: AdMeld and yield management
Before the invention of header bidding there was a different strategy for maximizing publisher revenue called “yield management.” While that phrase is still in use as a description of the publisher ad ops activities, it was once also considered a different class of ad tech vendor, related to, but not exactly the same as an SSP/Exchange.
The leading yield manager in the early era of programmatic was AdMeld, run by current Magnite CEO, Michael Barrett . The idea was to run a platform for the benefit of publishers, that would take in all the RTB and non-RTB demand and optimize it for the highest yield. AdMeld made the move to include RTB early, and gained a lot of share with major publishers.
In 2011 this all became moot when Google bought AdMeld and pretty much killed it.
AdMeld typically charged only a 7% revenue share compared to Google’s 20% revenue share on AdX. So rather than compete with AdMeld, what did Google do? It bought and buried it. --DOJ
EBDA aka Exchange Bidding aka Open Bidding
I’m going to restrain myself and not write five paragraphs about Google’s naming conventions. Let’s just refer to this as Open Bidding, the current name.
Open Bidding was launched by Google in 2018 after several years of testing, and the basic concept was to persuade the other exchanges to move from the header to a server-to-server integration. Google touted this system as reducing page latency and reducing setup complexity versus headers.
Open Bidding came with a number of catches that all benefitted Google:
领英推荐
Google intentionally designed this new form of integration [open bidding] to be just “slightly better” than early versions of header bidding. —DOJ, quoting a Google product manager
“Jedi Blue” brings Facebook into Open Bidding
Google’s biggest rival in advertising is Facebook — no other company comes close. When Facebook started aggressively capturing publisher inventory for its Facebook Audience Network (“FAN”) Google took notice. The nightmare scenario was if Facebook adopted header bidding and pumped its enormous demand footprint through that channel, making the AdX-GAM advantage much less vital as a revenue source to publishers.
Long story short, the two companies signed agreement that would let Facebook bid into Open Bidding under favorable terms. In exchange, Facebook would not enter the header bidding market — NOPE, NOPE, THAT DIDN’T HAPPEN, THERE WAS NO QUID PRO QUO ABSOLUTELY NOT — sorry, I meant Facebook logically concluded it was not worth their effort to enter the header bidding market.
Anyway, the deal was called Jedi Blue, it made a lot of headlines when it was revealed but in this author’s humble opinion was not a big deal.
“What Google wants: To kill header bidding” —DOJ, quoting an unnamed Facebook exec
Project Poirot
We haven’t spoken much about the buy-side of the Google stack. If you’re reading this newsletter you probably know that DV360 is the largest DSP in the market. In 2017, according to the DOJ complaint, more than 50% of DV360’s spend went to rival exchanges, and Google perceived this as both inefficient and a competitive weakness, since most of those dollars ultimately got redirected back into GAM through header bidding.
After experimenting with a plan to simply stop bidding into headers (didn’t work), the company developed Poirot, a system by which bids into headers were systematically reduced (by up to 90%). If you’re wondering whether the company actually called it Poirot, the answer is it had the amazingly anodyne name “Optimized Fixed CPM Bidding.”
Google’s argument for Poirot holds some water, so let’s give them equal time. At this point in programmatic’s evolution most auctions were still second price, and, as we all know, there was rampant price manipulation through soft floors and other shenanigans. A logical response would thus be to reduce bids to save money. The DOJ case rebuts this argument by pointing out that AdX also was not running an honest auction (they cite dynamic flooring) and if the DV360 folks wanted to benefit their customers they would similarly shave bids going to their own ad exchange.
Poirot ultimately moved DV360’s spend from ~40% on AdX to 70%, all at the cost of competitors.
“Immediately after the [Poirot] launch, OpenX experienced a 30% year-over-year decline in DV360 advertiser spend…and was forced to lay off approximately 100 employees.” —DOJ complaint
Unified pricing rules
Publishers using GAM were able to set different price floors for different exchanges. This could be done for quality reasons (charge more for lower quality ads), to compensate for business terms like volume guarantees, or whatever other reason a pub might have in mind. Often they were used to force AdX to bid higher than header bidding demand thus blunting Google’s advantages. Uh oh.
In March, 2019 Google announced “Unified Pricing Rules” which effectively deprecated this important feature in GAM and forced publishers to put AdX on an equal footing with other exchanges. The DOJ complaint is filled with evidence this was done cynically to help Google further consolidate demand, and that fig leafs were offered to irate publishers to no effect. These pricing rules further increased the proportion of DV360 buying that went to AdX.
Rather than me explaining how this played out, let my friend Stephanie Layser help:
“When we would raise floors on AdX, we could almost always make more money because their bids were modeled to pay just above what everyone else was paying, and they almost couldn’t decision without pricing data of all the rest of their competitors. Because of this, I would almost always recommend publishers floor AdWords differently than the rest, because their informational advantage drove prices down for pubs. They got rid of that ability during Unified Pricing rules, which was designed, like almost every one of these – to increase share of AdX and decrease everyone else, which just meant they gobbled up more inventory at lower prices. We saw a depression of yield just a few weeks after changes were implemented.” --Stephanie Layser, publisher ad tech exec
Epilogue
Header bidding certainly survived and thrived and is now the most common way exchanges integrate with publishers. Open Bidding appears to be a flop, with buyers like The Trade Desk refusing to work with it and many other DSPs “SPOing” it out of the supply path. But AdX gained enormous share along the way, and is not ceding that share even in the face of first priced auctions, government scrutiny, and wary publishers. Further, the DOJ excluded DV360 from the potential remedies in their complaint, so the actions on that platform are not likely to change.
This is the latest post of my new Marketecture newsletter: The Monopoly Report. Each week we will cover everything relating to Google’s ad tech antitrust trial. Subscribe for free here.
Principal Engineer & Industry/Community Liaison
2 个月Thanks Ari, your clear explications of shadowy operations and their impacts do us all a solid. These revelations also reassure me and I assume others who helped pioneer 3rd-party AdTech. Exposure of clandestine operations have enabled me to recognize why I had persistent feelings market dynamics were off back then. They explain why expected impacts from campaign changes didn't materialize, why invisible limits seemed to yoke our performance, why the results we saw when playing by accepted rules didn’t add up. We reviewed our systems and assumptions to no avail, what we could assess worked as expected. We wondered about manipulations, gamed them out, found patterns correlated with them, but ultimately put the efforts aside: our aspiring startup had more critical concerns. So we told ourselves no one would do such things. Even if they did, our only meaningful recourse was to adapt. It is discouraging to consider the losses that resulted from these antics. I’m hopeful we’re finally seeing a resolute shift in our approach to the web, one that affirms its preeminent potential for our world and future wellbeing and seeks to enshrine and protect it as a common asset that is co-owned by all of us for the benefit of all of us.
Building audience and revenue growth solutions for a privacy-conscious world
2 个月Great summary Ari. I only take a bit of exception to referring to efforts to help pubs find what buyers were willing to pay collectively as "rampant price manipulation". The challenge with the 2nd price auction model was that the floors often set the price, especially when so much demand was concentrated in just two bidders. The system effectively rewarded DSPs for concentrating bids, instead of encouraging more bid competition. Early flooring tests revealed buyers were willing to pay more, so it rewarded doing things like setting advertiser-specific floors. This didn't seem any different to me than regular negotiations on price that happen all the time between brands and publishers. I don't know what the ideal bidding and price setting model would have been, but dynamic price floors, for example, were just an attempt to try to find out a buyer's willingness to pay for an opportunity when the 2nd price auction model tended to benefit buyers, especially so as the buyside consolidated.
Founder @Venga!Labs - Repeat Entrepreneur - Angel Investor #entrepreneurship #startups #cx #saas #ecommerce #adtech
2 个月Olivier Jasmin
Scaling New Businesses | Rainmaker | Team Builder | Problem Solver
2 个月Great read and adtech history lesson.
Chief Technology Officer at Blis
2 个月Great wrap-up of the history and a reminder that not everyone in our ecosystem plays a fair game.