Shifting Attribution in Adobe Analytics
However, this has changed in recent releases of the Adobe Analytics product. Now you can apply a bunch of pre-set attribution models including J Curve, U Curve, Time Decay, etc… and you can also create your own custom attribution model that assigns some credit to first, some to last and the rest divided among the middle values. These different attribution models can be built into Calculated Metrics or applied on the fly in metric columns in Analysis Workspace (not available for all Adobe Analytics packages). This stuff is really cool! To learn more about this, check out this video by Trevor Paulsen from Adobe.
However, this post is not about the new Adobe Analytics attribution models. Instead, I wanted to take a step back and look at the bigger picture of attribution in Adobe Analytics. This is because I feel that the recently added Attribution IQ functionality is fundamentally changing how I have always thought about where and how Adobe performs attribution. Let me explain. As I mentioned above, for the past decade or more, Adobe Analytics attribution has been tied to eVars. sProps didn’t really even have attribution since their values weren’t persistent and generally didn’t work with Success Events. But what has changed in the past year, is that attribution has shifted to metrics instead of eVars. Today, instead of having a First Touch and Last Touch campaign code eVar, you can have one eVar (or sProp – more on that later) that captures campaign codes and then choose the attribution (First or Last Touch) in whatever metric you care about. For example, if you want to see First Touch Orders vs. Last Touch Orders, instead of breaking down two eVars by each other like this…
…you can use one eVar and create two different Order metric columns with different attribution models to see the differences:
In fact, you could have metric columns for all available attribution models (and even create Calculated Metrics to divide them by each other) as shown here:
In addition, the new attribution models work with sProps as well. Even though sProp values don’t persist, you can use them with Success Events in Analysis Workspace and then apply attribution models to those metrics. This means that the difference between eVars and sProps is narrowing due to the new attribution model functionality.
To prove this, here is an Analysis Workspace table based upon an eVar…
…and here is the same table based upon an sProp:
What Does This Mean?
So, what does this mean for you? I think this changes a few things in significant ways:
- Different Paradigm for Attribution – You are going to have to help your Adobe Analytics users understand that attribution (First, Last Touch) is no longer something that is part of the implementation, but rather, something that they are empowered to create. I recommend that you educate your users on how to apply attribution models to metrics and what each model means. You will want to avoid “analysis paralysis” for your users, so you may want to suggest which model you think makes the most sense for each data dimension.
- Different Approach to Implementation – The shift in attribution from eVars to metrics means that you no longer have to use multiple eVars to see different attribution models. Also, the fact that you can see success event attribution for sProps means that you can also use sProps if you are using Analysis Workspace.
- sProps Are Not Dead! – While I have been on record saying that outside of Pathing, sProps are just a relic of old Omniture days, but as stated above, the new attribution modeling feature is helping make them useful again! sProps can now be used almost like eVars, which gives you more variables. Plus, they have Pathing that is better than eVars in Flow reports (until the instances bug is fixed!). Eventually, I assume all eVars and sProps will merge and simply be “dimensions,” but for now, you just got about 50 more variables!
- Create Popular Metric/Attribution Combinations – I suggest that you identify your most important metrics and create different versions of them for the relevant attribution models and share those out so your users can easily access them. You may want to use tags as I suggested in this post.