Have Google in one fell swoop killed the digital attribution market?

GDPR has not even landed (25th May 2018 – put the date in your diaries) and we’re already seeing the Law of Unintended Consequences at work.

https://adexchanger.com/platforms/google-sharply-limits-doubleclick-id-use-citing-gdpr/

What is digital attribution?

It’s the process of using individual event-level data, creating paths for individuals path to purchase (or some other key event) and then assigning a value to every step along the path to that event.
It relies heavily on data being available at the individual basis. DoubleClick log files have long been a common source of this data. In a previous life I passed many a happy hour playing with the logs.

Why is attribution needed?

Other solutions are not satisfactory, either in their ability to get to a true picture of which marketing levers are really doing the work (last-click attribution), or in the granularity of analysis provided (econometrics).

What have Google done?

Google will no longer include the DoubleClick ID in the log files, this is the key field that allows you to take individual events and create a picture of the individuals path to purchase.

Why have they done it?

GDPR. At least that’s what they say. Although cynics and long-term Google-watchers are tempted to question whether their motives are quite so pure. After all, Google offer their own attribution service through the husk of Adometry which they swallowed whole a few years ago.

Should I care?

Good question. Probably not. Well maybe for some, but is that you? Our viewpoint is three-fold:

Do you know the incremental effect of each channel?

Traditional analytical techniques such as econometrics can tell you the true ROI at the channel level (and sometimes below this). This is your starting point before getting down in the weeds of digital attribution modelling.

Is your log file data any good?

Do you have natural search tracked? Are you looking at visibility? Facebook impressions aren’t tracked. There is no traditional ATL media in there. There is almost no data unification for users with multiple devices (i.e. everyone).

How many steps do you actually have?

Have a look at your data. What percentage of your customers only have one step in their path to purchase? 70%? 80%? We have seen data sets in which over 90% of purchases took place after a single recorded step.

No amount of clever attribution analysis is going to change the picture from last click if the majority of your recorded paths are just one step. That’s not to say that your consumers actually have a single-step path to purchase, just that your data tracking is incomplete, for any number of reasons (cookie blocking being high on the list).

I need my digital attribution, what can I do!?

Panic not! There are still options open to you.

Use a provider who generate their own data

Generated via their own pixel-based IDs. Big providers already offer this solution, but it does mean retagging your landing sites.

Use Google

If you already serve all your media through Google, then you could in theory just use their own attribution platform. That is, if you trust them to both serve your media AND evaluate its impact. Some may see a conflict of interest in that mix…

Use a provider who works with Googles Ads Data Hub

This allows people to recreate their approach in the Google Cloud, and access more detailed (but still not everything) log files which would allow the attribution to be done.

Shock! Horror! Use another ad server

Yes, they DO exist. Not a friction-free solution, but certainly an option and I’m sure they will be queueing up at media agencies to offer their services in the light of this change.

So, what should I do?

What do you currently know, and what do you want to know?

If you don’t know the true ROI of your digital channels, an econometric model (or a series of potentially nested models) could, and potentially should, be your first port of call. For those who already know their display ROI (e.g. through econometric evaluation), but have a burning question of DMP vs. PMP, we have had a lot of success in using econometric models combined with well-structured testing programs.

Get in touch and we can tell you more.

Sam Watts

Sign Up
*