I tried to understand header bidding in programmatic advertising, but I hit a brick wall of incomprehension.
I got completely confused reading about how real-time bidding (RTB) works.
I understand the theory and the different components of the RTB ecosystem, but after reading countless articles it seems that the reality is a complete mess.
And a thoroughly Kafkaesque mess at that. Okay, this isn’t exactly news, but I thought that rather than write about header bidding, it might be more interesting to write the story of a layman trying to understand it all.
This much I knew
Real-time bidding is the process of selling advertising space via an open ad exchange auction.
These open auctions are where publishers often sell their less valuable inventory, after the more sought after spots have been sold directly to preferred advertisers (through programmatic direct, private marketplaces or preferred deal).
For a simple explanation of real-time bidding and programmatic direct, you can check out Christopher Ratcliff’s Beginners Guide.
Essentially, RTB allows advertisers to bid on different users (from pre-defined demographics) that pitch up at different types of web content. Successful bidders pay the second highest price (so they don’t lose out if they bid far more than everyone else) and their adverts are served to the user.
This much, I knew. And it made sense to me.
Don’t go chasing waterfalls
But when I tried to read up on the inefficiencies of RTB (which led to the development of header bidding), things started to get confusing.
Lots of people seemed to mention an inefficient waterfall process, whereby unsold inventory is offered by the ad server to one ad exchange after another, the top-ranked one first (estimating what the exchange can bring in), then passed down to lower ranked exchanges if the inventory is not sold.
This ranking of so-called ‘demand sources’ is apparently one of the problems with traditional RTB. This ranking doesn’t take real-time demand into account, so the low-ranked exchanges, though perhaps smaller, may have stumped up a higher bid (and so publishers lose out).
Google’s Doubleclick for Publishers is a commonly used ad server, which gives priority to (surprise, surprise) Google’s ad exchange, AdX, which can pip other bidders.
Conceptually, I understand this, but then I thought, what about supplier side platforms (SSPs), where do they come in?
But what about SSPs?
It was my understanding that SSPs were designed to offer publisher inventory to multiple ad exchanges and provide tools with which they can optimise their yield.
SSPs are meant to make everything easier. So, why all this waterfalling and inefficiency?
I delved into definitions and more theories and found that publishers not only have waterfalls of ad exchanges, but they waterfall SSPs, too.
This seems like complicated madness. Publishers push inventory through their first SSP at a high price, dropping the price if it doesn’t sell and pushing through the next SSP.
Aside from the publisher-side headache of cascading bids through these multiple SSPs (which were supposed to simplify things), there’s the slightly dubious idea that the exact same inventory is priced differently depending on where an advertiser buys it.
If I’m honest, at this point I was still a bit confused about what was an SSP and what was an ad exchange. SSPs are routinely referred to as ‘demand sources’ (I thought they were for selling?) and I had to do plenty of Googling to find out how certain platforms are described.
As those in the know will tell you, there is now little differentiation between SSPs and ad exchanges. They both aggregate inventory, but the difference is that SSPs should put more focus on publisher value.
However, each adtech platform out there has sought to add functionality, meaning that SSPs and ad exchanges are almost the same thing. Ad exchanges included SSP functionality – Google’s ad exchange did this back in 2011 (rolling SSP Admeld into its platform).
So, the bottom line is that SSPs are just like ad exchanges, and they don’t offer complete access to the market, because all the different SSPs are competing and want to protect their demand sources, which fragments everything.
Hang on, you still haven’t told us what header bidding is!?
So, all the bids come in at the same time and this can mean a higher price is achieved. The publishers can even let the open exchange beat a direct-sold impression, if the price fits.
There’s a debate about whether header bidding decreases or increases latency – its advocates say that because all bids come in at the same time, there’s less chance of a time-out (when the publisher would fill the spot with its own internal ads).
However, others think header bidding increases complexity and also latency (notably Google, which doesn’t exactly stand to benefit from header bidding, given it breaks the dominance of AdX).
Google is fighting back though
Google has rolled out First Look as its own attempt to solve the same problem as header bidding does.
Publishers using the Doubleclick ad server can use its Dynamic Allocation tool to see in real-time how much media buyers within the Google ecosystem are willing to pay for an impression.
What’s my overall feeling after such a nightmare?
SSPs should talk to each other, but they don’t and that’s a problem for publishers who want to make the most of their inventory.
Some very brief thoughts on my nightmare:
- Why aren’t adtech companies better at explaining what they are and how the process works? Perhaps they don’t want to confuse people.
- I realise that the internet isn’t really a democracy, but until some of these processes are replaced by a more elegant solution, doesn’t the complexity mean that bad practices take longer to come to light?
- How can we expect fraud and viewability to be tackled without greater integration?
- There’s a skills shortage in programmatic – the industry needs to get better at explaining all this stuff.
- No wonder publishers are investigating bespoke and native formats.
A big caveat: My knowledge is limited. That’s the whole point of this article. If, reader, you find any errors in what I have written above, please set me straight in the comments below.
For more on this topic, check out these Econsultancy resources: