“Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Eric Berry, co-founder and CEO at TripleLift.
Back in the early days of the web, it was fairly intuitive that every publisher couldn’t work with every advertiser. As a result, we saw the creation of ad networks, remnant fill, daisy chaining and the like – one seemingly good innovation after another that led to a fragmented world of high latency and suboptimal monetization.
Ultimately, to right all these well-intentioned wrongs, the ad exchange was born. Everyone could bid all at once and the maximum bid – subject to the publisher’s restrictions – would triumph. With no daisy chaining and the highest conceivable RPM, everyone would win, right? Wrong.
The world of display advertising took a strange turn. The advent of the ad exchange, and the ability to value ads on a per-impression basis, led the ecosystem to accept that per-impression valuation was an appropriate and sensible goal. Since banner ads are most effective on a direct-response basis, as the reasoning goes, we should take this direct-response valuation to the extreme and create the highest level of transactional efficiency with incredibly sophisticated bidding algorithms that are able to eek out every last bit of ROI.
This is a noble goal, but it’s just not realistic.
Ad Exchanges Vs. Financial Exchanges
Ad exchanges are not like financial exchanges. The stock market is a constant matchmaker in which a completely identical asset is traded over time. A share of GE common stock is always the same thing, whether you buy it from seller X on Monday morning or seller Y on Saturday night. It always represents one unit of equity ownership and always comes with the same bundle of rights. Thus, it can be precisely valued based on the underlying nature of GE and one’s own guesses about the future of GE and any other trends.
This is very different from ad exchanges. When you buy an impression on an ad exchange, you have almost no idea what you’re getting. You may learn if the ad is subsequently viewable, but you never know if it’s actually seen. You also don’t know the user’s mindset, if he or she cares about the brand or even if the person using the computer is the same consumer you’re targeting. Furthermore, since most users never click on ads and because attribution is generally very simplistic and inaccurate, you also don’t know if your ad had the actual impact you wanted – possibly until weeks or months later.
It’s not that there isn’t some room for optimization, but it is very clearly the case that the idiosyncratic dynamics of automated ad buying preclude the ability to truly and confidently value the underlying asset being bought on a per-impression basis. This means that, at best, ads can be purchased and optimized on a statistical, large-scale basis. But this must be considered in light of the fact that open exchanges generally include unknown counterparties – meaning unknown websites and users – both of which could be malicious actors.
The Downside Of Open Exchanges
The Internet is a complex adaptive system in which agents create sophisticated strategies to maximize their own income through potential inefficiencies or loopholes. The problem with open ad exchanges is that ads simply cannot be accurately valued on a per-impression basis. Yet they are sold as if they could be, creating a huge market for exploitation. It was only a matter of time until this was discovered, and the problem has now grown so rampant that on some video exchanges, fraud may account for half of all impressions transacted.
There is not, nor will there ever truly be, an effective way to determine if an individual impression of a single ad actually had an impact – or to measure what that impact was. This means that statistical analysis will be the modality for valuation going forward. If ads are to be sold in a manner that allows a comparisons of individual buyers’ valuations – which they should be – then we must make fundamental changes to how ad exchanges work.
Every seller must be clearly identified. This goes beyond the name of the exchange and must include the corporate entity that is ultimately being paid for the impression. There is material counterparty risk in online advertising, so it must be evident who the counterparty is so that buyers can make appropriate determinations. Every step of the value chain must be evident and all bad actors – and even unproved actors – must be eliminated or vetted.
Further, every impression must be clearly identified. Exchanges should share a common protocol for unique impression identifiers and include them on pass-backs and other redirects. This will allow buyers to more accurately determine the true provenance of each impression all the way to the original seller.
Exchanges must also proactively deter fraud. Every exchange should partner with independent third-party vendors that apply statistical analyses to determine fraudulent traffic. Buyers must be credited against the fraudulent traffic by the exchange. In turn, the exchange would be incentivized to ensure that publishers reimburse them for fraud, and so on. This would instill a higher level of vigilance throughout the ecosystem, rather than the current, unsustainable model of caveat emptor.
Finally, the industry should perhaps admit that per-impression valuation is useful primarily to allow for a comparison of valuations. It is true that demographic, contextual and other analyses feed nicely into RTB valuations and do increase the performance, but the trade-off is that when buying from an unscrupulous partner, the fanciest machine-learning algorithms translate into exquisitely valued bot traffic.
Buyers would be wise to re-evaluate the breadth of their exchange targeting to only include trusted providers that actively monitor and vouch for their own traffic, and to increasingly explore programmatic deals and private marketplaces with trusted partners.