Welcome back to our series on attribution in digital advertising. Last time around we talked about the basics of attribution, what it is, and the challenges around the most simple methods of attribution. Today we're talking about some more advance methods of attribution and some ways to go beyond simple last-click and last-touch attribution.
Building a Multi-Touch Attribution Model
We briefly mentioned multi-touch attribution last time around, and how you can add things like Time Decay to add more or less credit depending on when an action occurred in a consumer's purchase journey. But how do you create those weights without just setting arbitrary numbers? And how do you create better models? The answer is data - your adserver and other platforms can provide important data to help you understand how people are arriving at your site and what they're doing, while Google Analytics can be helpful for understanding their behavior once they arrive.
You can use the data from these two sources - Google Analytics and Google Ads - to build a clear picture of how your customers are likely to end up on your site and make a purchase. And it's worth doing the work at this step to identify common intermediate steps and determine what impact, if any, they're having. If you're seeing the same conversion rates or chance of conversion from people who see a social ad before clicking as people who don't, that ad or campaign may not be adding much in its current state/execution. Likewise, you may find that people aren't clicking on display ads, but those show up in every successful conversion purchase journey, indicating that they're a key part of driving awareness even without a click. it's important to examine what these factors are and use those to determine your weights.
Data-Driven Attribution
If that sounds complicated, the good news is that there are already tools out there which can make it easy to implement - if you're running campaigns large enough to have at least 300 conversions and 3,000 clicks/ad interactions per month, then Google's built in Data-Driven Attribution model can be put to use to give you a dynamic attribution model which uses your data to optimize bidding and delivery based on the performance of your keywords, ads, ad groups, and campaigns. Note that making good use out of this tool requires having multiple campaigns and formats in the field and running them through Google Analytics/Ads - if you don't have enough data you'll just get the same results as last-touch attribution, though that isn't the end of the world.
If that still sounds complicated well, the good news is that you can give us a call and we'll be happy to set it up for you.
Getting Weird with It
Data-driven Attribution is just the start, and you don't have to use Google's tools for it; you can build your own multi-touch attribution modeling in a number of ways. One way to do this is with Markov Chains. When you're building a Markov Chain model you essentially plot out each potential node in a customer journey, treating an ad exposure or a visit as a node on a chain. Then you count the number of times a customer went from that node to each other node and use this to build a probability graph which can both show the relative value of each node and how the model would have been affected by removing a node. This can then give you relative weights to use for each touchpoint in your model.
When you have the data to do it, this is absolutely the best approach, but it's also complicated and not necessarily going to matter for every marketer and purchase journey.
Cookieless Attribution and CTV
Savvier readers will point out that all of this attribution tends to rely heavily on cookie-based tracking. What do you do if you don't have access to cookies, either because you're using cookieless channels like CTV and mobile, or because your consumers tend to disable cookies? Well in those cases it becomes much more difficult to properly track exposure, but it's not an impossible problem to solve. The solution there is often custom technology used to track consumers cross-device, creating a "device graph" for a given person or household that connects a large number of devices to a specific ID or IP address. This then allows for graph-level attribution, which solves most of your problems. There are some accuracy issues that crop up when you're combining devices, such as shared screens in a household, but in many cases treating a household as your target rather than a person in that household is a better strategy anyways - larger purchases like cars or furniture or travel campaigns typically benefit from targeting multiple people in the same household to influence a purchase or decision.
Next Time: Incremental Impact
That wraps up this week's post but next time around we'll talk about measuring incremental impact as an alternative to attribution models, how that works, and how you can implement it on your campaigns.
Interested in talking about or implementing a better attribution strategy? Drop us a note in our Contact form and let us know!
Comments