We often get asked to explain our attribution modeling algorithm. Our response: “What attribution model would you like to use?” RJMetrics doesn’t use one specific algorithm; rather, we can support whatever algorithm our customers would like to use…provided the data exists in your database or in one of the many marketing tools we integrate with.
The question we get asked then is: which algorithm should we use? Last click? First click? Some mixture? And how does user source vs. order source tie into it?
My advice to our early stage customers is this: keep it simple. When you’re young, you likely don’t have many users and transactions. Even if you do, you don’t have the benefit of time to understand what the impact of pulling different marketing levers has for your business.
So what does “simple” mean? Simple doesn’t have to mean true last click, or even google-analytics-sort-of-last-click. It doesn’t have to mean equal apportionment across a stack of source paths either. Simple means striving for end-to-end completeness rather than algorithmic complexity or behavioral nuances. And perhaps most importantly, “simple” means framing your quest to define an attribution model in the context of a real business or marketing decision. Whatever time you spend on attribution modeling had better be directed at spending your marketing dollars better; if you’re just trying to satisfy an OCD need to perfectly count all the beans, stop.
It is probably worth mentioning there are two kinds of attribution. The first is path attribution, i.e. the user came to our site from source A, then source B, and then source C before finally doing “interesting thing X”. The other is transactional attribution, i.e. the user came to us from (last? first?) source A and then did interesting thing X, then came to us from (last? first?) source B and did interesting thing Y, and so on.
Another way to say it is that path attribution is more focused on the pancake stack of user behavior before they did something interesting; and transactional attribution is more focused on apportioning a succession of interesting acts (usually revenue events) to a set of one or more sources. These aren’t mutually exclusive – but they are often best looked at using different tools, especially by budget-conscious start-ups.
Here’s an attribution modeling process our Client Analytics Services team has found useful for startups:
Step 1: Capture all the data
Capture source not just for the user or the first purchase, but also each follow-on purchase. Which source? Start with what’s in your GA cookie. Read our article about how to store that information into your database here. With this in place, you can properly perform transactional attribution modeling using your database. You can give that user’s source credit for purchases, and give the first purchase source credit for follow-on purchases, and so on.
Step 2: Give credit where due
We find that using repeat purchase probability often works well. If you know how hard it is to get someone from order one to order two, you can appropriately give some credit for order two to order one’s source. On the other hand, if getting someone from order three to order four is pretty likely, then there’s not much need to give source-based credit for order four—rather, it’s the fact that the entirety of the company experience for the customer has been good that is driving orders, not the source itself. (If you don’t know your repeat purchase probabilty, stop reading this blog post and go figure it out. We can help!)
Step 3: Refine using path attribution
While there are lots of sophisticated tools for this, we find that Google Analytics’ enhanced attribution modeling is more than up to the task for startups. As a sanity check on the transactional attribution you did in step 2 above, take a look at what sources are on the path to a given purchase. Take the top three to five of those, match up against your top hard-to-get orders (usually orders one and two), and fine-tune your model.
Using the basic steps outlined here, we find that two important things happen: marketers are able to make decisions based on not just initial conversion value but lifetime value, and proper credit is given for how challenging a particular order in a user’s lifecycle is to get. Just doing these two things should increase your ROI and optimize your resource utilization.
If you would like help specific to your business’s attribution model, talk with the team of marketing experts on our Client Analytics Services team today by contacting your RJMetrics Account Manager. Don’t have one? Let’s fix that.
Data-driven tips and how-to’s that help your business go from 0 to 60.