In part one of this series, VAN traced back the origins of some of the most important components of the digital video advertising ecosystem – video ad serving, VAST, the ad exchange and the supply-side platform.
As we said in part one, while we try our best to verify claims, they are mostly anecdotal. The article sparked a bit of lively debate as several people came forward to dispute the accounts given, making their own case for being the original inventor of some of the bits of tech listed.
Now in part two, we’re looking at the origins of the publisher-side ad server, the demand-side platform, the video exchange and header bidding.
Publisher-Side Ad Server
In part one, we looked at the creation of the first video ad server, a title claimed by Tom Des Jardins at LightningCast. This week however we’re going back even further to the inception of the very first publisher-side ad server, created back in 1995 by Tom Shields and John Danner at a company called NetGravity.
Shields says he didn’t set out to create an ad server (arguably planting the seed for the creation of the wider ad tech ecosystem in the process), but rather he and Danner were looking for interesting problems to solve, and spotted an opportunity on the early internet.
Digital publishers did run ads at the time, but they were manually added as .jpg images on web pages, with the number of clicks that web page received being the only real form of measurement available for the ad campaign. Shields and Danner were hired as consultants by Yahoo, at the time a very small company, to write a script which would automate this process.
It soon became apparent that this tech opened up exciting new possibilities; Shields recalled Yahoo’s excitement at the realisation they would now be able to rotate ads as often as twice a day!
Through word of mouth, Shields and Danner were connected with Time Warner. “They were using the open market web server, which was built in a language called TCL. We re-wrote the ad server script in TCL so it could run inside the web server, and that meant that ads could be switched depending on who was asking, which was a brand new idea at the time.”
Time Warner’s ad server went live around Thanksgiving in 1995 – Shields says he had to skip Thanksgiving with his family to debug the script. Shields and Danner at the point understood the commercial potential for their product, and began working on a commercial version in 1996 having started their company, NetGravity.
While publishers were interested in NetGravity’s product, Shields says it was hard to say at the time how much it would take off. Many advertisers were unfamiliar both with the idea of banner ads and of one-to-one targeted advertising, meaning only the more forward thinking advertisers were interested in running digital campaigns. Plus, the Internet itself at the time wasn’t firmly established, and many believed interactive TV was more likely to gain traction.
It was hard to take full advantage of what the ad server was capable of at first. For example, Shields explains, the ability to target individuals was seen as valuable, but it was also difficult to execute due to the lack of data available. Also, in many cases measurement would have only served to highlight how small the audiences were due to the relatively low adoption rate of the Internet at the time.
Over time though of course, the market continued to grow, and in 1999, NetGravity was sold to DoubleClick in a $530 million stock buy-out. Shields says there was no one moment when the potential for how big digital ad serving could become became apparent, rather it’s been a gradual process of realisation over time. “I remember there was a time when we thought, ‘wow, I wonder if we’ll ever serve one million ads in a day,’” he said. By 2012, Google was serving an estimated 29.8 billion ads per day.
DSP
Those VAN spoke to while tracking down the origins of ad tech agreed that it’s hard to point to one specific company or individual for the first demand-side platform (DSP) as different people use different definitions for what a DSP actually is. MediaMath’s CEO Joe Zawadzki acknowledges that the definition has evolved over time, but stakes his claim to having created the first one.
Zawadzki started at a company called Poindexter Systems, later renamed to x+1, which handled website optimisation, helping retailers and advertisers use data and analytics to their customers’ experience through personalisation.
Zawadzki heard from clients that while they liked the website optimisation tech, their core problem was getting the right customers onto their sites in the first place, so this is where the business’ focus shifted to.
“Had I known that was going to begin a twenty year journey of trying to figure out the media world… well, I would have done it anyway because I’m a glutton for punishment,” joked Zawadzki.
At the time advertisers were mostly just rating digital publications based on their audiences in the same way they would rate magazine or TV shows, in order to target potential customers, but this was in many cases very inefficient. Zawadzki gave the example of American Express trying the run ads for its Exclusive ‘Centurion’ card – even on a site like the New York Times, around 98 percent of spend would be wasted as the people the ads were shown to would be ineligible, or perhaps would already own a centurion account.
Zawadzki and his team had several cracks at trying to solve this problem, the last of which was bulk buying impressions and allocating them across clients in a transparent way with fixed margins, which he described as a ‘proto-DSP’. But the market at the time was somewhat unprepared for these sorts of products, with only the “bleeding edge” clients with specific requirements showing real interest.
However 2007 brought with it what Zawadzki describes as “a moment of punctuated equilibrium where lots of things happen at once: the meteor strikes, the dinosaurs die, and you have this whole accelerated phase of new species emerging.”
A series of big acquisitions took place, including Google’s purchase of DoubleClick and Yahoo’s purchase or Right Media, signalling where these companies were looking for growth. At the same time, the first ad exchanges had opened up shop and the first data exchanges like BlueKai were emerging, allowing for the use of third party data for targeting.
During this time Zawadzki had left x+1 to start MediaMath alongside Greg Williams and Eric Wasserman, and the three saw an opportunity to capitalise on these transformations that had hit ad land.
“We said there needs to be a platform to stitch all of this together – you’ve got to stitch all of those media touchpoints together, and connect to those automated media exchanges, you’ve got to be able to layer in all the 3rd party data that was becoming available and marry it to the advertiser’s first party data, and then in order to make sense of those ingredients, you need machine learning in order to make those decisions in the 50ms you have to respond,” said Zawadzki.
MediaMath signed up big clients including Citibank and AOL, which helped the company sell its product elsewhere, and helped the formation of the first internal trading desks at agencies to help reach a wide range of advertisers without having to contact them individually.
As the tech took off, what MediaMath first called its ‘marketing operating system’ became known as the DSP, though the definition became more narrow over time, referring only to the media execution aspect of the platform, according to Zawadzki. Now however as the market evolves DSPs are evolving back into his original vision, with the data and AI aspects woven in, he says.
Video Exchange
VAN in part one traced the birth of the ad exchange and real time buying (RTB) back to Brian O’Kelley at Right Media (though Jason Knapp, now vice president at Viasat, has since said he and Fabrizio Blanco have a decent claim to inventing it at Strategic Data Corp).
The video exchange meanwhile had its own separate evolution. Adap.tv claimed to be the first to bring the ad exchange model to the video world when it launched ‘atm’ back in 2010, but SpotX co-founders Mike Shehan and Steve Swoboda say they were working on their video exchange as early at 2006.
Shehan and Swoboda started out at their company Booyah Networks building a pay-per-click search engine, alongside other early competitors like Overture. “As you know, Google pretty much won that battle,” said Shehan.
The two however had enjoyed working on the model they had built at Booyah, and looked at where else they could apply the same sorts of auction dynamics that they’d used for their search engine, eventually settling upon video advertising.
“Video back then was around a $200 million dollar industry,” said Shehan. There was no dynamic ad insertion, and video ads were by and large sold through direct deals with publishers, with advertisers receiving an excel spreadsheet at the end of the campaign detailing how many times the video the ad was attached to had been played.
Shehan and Swoboda began working on their new product, and in April 2006 put out a press release announcing the launch of “an auction-based video ad serving platform”.
While display exchanges were also emerging at the time, Shehan says video had its own, somewhat separate evolution. “We developed SpotX having zero knowledge of the display ecosystem, so I think it’s pretty interesting how display and video have developed, where there are a lot of similarities but there are also a lot of differences,” he said.
“For video, there was nothing like that around at all at the time, and for three years it was painful,” said Shehan, who recalled pitching to agencies who at the time weren’t particularly interested in what he and Swoboda had to offer. “There was no ecosystem for what we’d created,” said Shehan.
SpotX, then SpotXchange, was at the time a division of Booyah but split off in 2007 to become its own company, dedicated to video advertising, and though it took a while for the buy side to come around to the idea, the growth of video focussed companies like DataXu and TubeMogul helped push the industry forward.
“In 2010, our business started to explode, and that was a result of programmatic video finally taking off and agencies buying into the idea of dynamically inserted ads,” said Shehan, and it was in this year that SpotX released its real-time bidding functionality for its platform.
The company continued to update its platform, adding its SSP and ad server tools in 2012/13, and moving into the connected TV and over-the-top space later on, rebranding to SpotX to reflect its move away from being only an ad exchange, and has since been acquired by European broadcaster RTL Group.
Header Bidding
The inventor of header bidding is contested, perhaps partly due to the fact that it came under several different guises in its early years. OpenX is credited as it’s inventor in a Campaign report, and OpenX’s CRO Jason Fairchild has previously referred to the tech as “a product of ours.”
Beeswax’s Ari Paparo however backs up Brian O’Kelley’s claim that he invented header bidding back in 2009, which AppNexus called ‘pre-bid’. O’Kelley says the tech was a direct response to a DoubleClick for Publishers feature called ‘Dynamic Allocation’.
DFP, now owned by Google, allowed Google’s own ad exchange AdX to compete against publishers’ directly sold campaigns, while other exchanges were only able to compete for impressions not bought either through direct campaigns or AdX.
Header bidding was developed as “a hack around a feature gap in DFP”, says O’Kelley. AppNexus’ pre-bid tech was developed to run a client-side auction which puts out a bid request to one or more ad exchanges and networks, with the winning bid then being sent through to DFP. If this bid then won when competed against direct campaigns through DFP, a javascript function called up the ad from the winning network or exchange.
Header bidding’s use spanned beyond simply countering DFP though, as it emerged as a popular alternative to the ‘waterfall’ system which offers impressions to demand sources one after the other. Google has since updated DFP too with it’s own take on header bidding, called exchange bidding, which serves a similar function.
Others however came upon the same idea through different route. Lee Cassingham said that his company Viewex came up with their own header bidding solution as part of a viewability product.
Cassingham, having run various businesses within ad tech including an automated programmatic revenue /performance reporting tool, a viewability analytics tool, and an ad ops consultancy, looked for ways to create more value on a publishers’ page, and created a 100 percent viewable ad format which appended itself to the bottom of a page.
This then inspired a different solution to the viewability problem, where Viewex sought to build a capability for its own ad server whereby it could make viewability-based decisions on a web page before an ad loads. Whilst building the technology Cassingham realised he could capitalise on this pause by adding another step to call a third party to prospect for a higher bid and increase revenues
To measure viewability, Viewex needed to put a wrapper containing its viewability tech around the ad unit, which required placing a script in the head of the page, which the company called header injection. The tech pauses the loading of an ad, verifies the viewability of the unit, and calls Viewex’s server for assets, while also calling third-party platforms to prospect for higher bids
Cassingham doesn’t claim to have invented header bidding (Viewex’s product came out in 2013), but his story provides an interesting example of how the same sort of technology can evolve independently while being built for completely different purposes.
He does, however, believe the open source header bidding tech that has emerged needs a rethink. “Fundamentally many of the challenges faced today in header bidding are related to the features of the technology being designed for a slightly different set of user requirements,” he said. “Unlike today, we didn’t expose publisher user data with every request so there was little risk of revenue cannibalisation, server load was lower so it was cheaper to run, we had machine learning processes in place to minimise manual optimisation and predictive methods to minimise latency. The technology was open sourced too early without enough consideration of risk. Inertia across the industry is creating problems and the core header bidding technology needs rethinking from the ground up.”
