Five Inconvenient Truths of Multi-Touch Attribution (MTA)
After several years of development and heightened industry expectations, the MTA bubble is bursting before our eyes. With the sun-setting of Convertro by Oath, the internalization of Adometry by Google 360, and the recent purchase (and likely absorption) of Visual IQ by Nielsen, the once big three of MTA have now all but disappeared. Why?
On the surface, these large media networks saw a chance to bring in-house a technology to enhance and optimize ad-spend within their own programmatic platforms, but the truth is that MTA as a stand-alone service failed to deliver and failed to live up to the industry’s expectations.
That said, there is still an assumption today that MTA is the holy grail for measuring the impact and details of ads. For example, in Martin Kihn’s (Research VP, Gartner) recent analysis of Google’s decision to no longer make their DoubleClick IDs available in its ad server log files – a foundational element for any MTA solution – that AdExchanger recently published, he still asserts that “MTA is not the only way to measure the true impact of ads, but is theoretically the most accurate and provides by far the most detailed results.” However, we are far from convinced of MTA’s superiority as a measurement tool based upon experiences numerous companies have had with it during the past 10 years that MTA has been around.
Here are five inconvenient truths about MTA that also explain what is behind its missed opportunity. The following points are not theoretical. They are results of first-hand observations and experiences developing and deploying MTA solutions. They are also based on numerous companies’ feedback, experiences, and their attempts to implement solutions from virtually all of the major vendors in the MTA space.
MTA was not a measurement of incremental impact, and therefore cannot be “true” ROI.
If you take a slice of time, you can only understand the relative performance of placements in that period, not the absolute performance (i.e., true ROI) since the analysis by definition cannot control for seasonality and the many other environmental drivers that are influencing response.
MTA was not at the customer level.
MTA was positioned as the ultimate attribution and marketing ROI measure at the customer level when in reality it was simply a better currency than Last Touch Attribution (LTA) at the cookie level. MTA was further sold as a ‘step up from MMM,’ allegedly providing a direct attribution of sales.
The truth is it was never at the customer level, and one individual can be associated with many cookies. Even with custom tag management, the match-rates across the multiple cookies served on the many devices we use every day yield shockingly low deterministic match samples or pointlessly broad probabilistic matches.
MTA was not real-time.
MTA was understood to be real-time when, in fact, it is at best near real-time. Modeling too soon will miss responses to the original stimuli that haven’t occurred or been captured yet, and premature measurement will lead to undercounted responses and ultimately miss-attribution.
MTA was unable to inform the media buy.
Regardless of the ability of MTA to scale to real-time, the ability to change a display buy was heavily restricted since deals were locked in weeks ahead under traditional direct buys. Furthermore, with the arrival of programmatic buying, these opportunities are even fewer now.
MTA had impossible data requirements.
The data was never there and never will be. The hard truth is that there is no unified data set at the consumer level that is tracking all exposures and all behaviors and transactions, nor has this kind of data been available. A minimum of 3 years of data would be needed to create a truly unified MTA model that controls for all drivers of transactions and makes a true assessment of incrementality and ROI. Anything short of 3 years will generate bias and therefore miss-estimation.
Particularly given recent issues around privacy we can only imagine this type of complete dataset is a very long way off. But, even if it were available, the MTA methodologies used are all inappropriate for such a high dimensional time-series data set.
Why does all this matter? It’s important to understand and agree on these facts as an industry. If the fast adoption of new approaches are to gain support, then it’s also important that we as an industry quickly and earnestly learn from our mistakes. In order to resolve and advance the quest for solid data and solid analytics that rigorously demonstrate and support marketing, it is imperative that we are all aligned on what is, and what is not, possible and why.