March 17, 2020
Facebook App Events: Why They’re Important and Which Ones to Track
“Which attribution model should we use?”
This is a question that we get from almost every advertiser that we work with, sometimes across their entire media mix, but especially over the attribution model within Facebook Ads. While there is no “right” answer and no best practice that shows you what you should use, we adamantly believe that the best attribution model is the one that everyone agrees on.
To that end, a quick note on attribution in general before we discuss scenarios for Facebook. No attribution model will be perfect or complete. So the idea behind attribution choices shouldn’t be trying to determine the exact value of every penny you spend, but instead to use realistic and reasonable assumptions to approximate the value of a given marketing tactic. As an example, position-based attribution has been popular with digital marketers over the last several years. The values given to each position on the conversion path are never exact, but they aren’t random or arbitrary, either. They’re based on an agreed-upon and realistic point of view on how each position contributes to a given action.
Okay, with that out of the way, which Facebook attribution model should you use? Start by considering these two items:
If your source of truth is something like Google Analytics or a CRM system, we usually recommend using a 28-day click model for Facebook performance evaluation. Neither GA nor your CRM has the functionality to give credit to Facebook impressions via view-through actions, and you’ll want to make sure your in-channel data set matches the source of truth as closely as possible (in all likelihood, it still won’t be exactly the same, and that’s fine). This is just for apples-to-apples performance comparison. To be clear, we would still recommend you continue to incorporate view-through conversions as a directional metric, especially in prospecting audiences.
For advertisers that are using channel or platform data as a source of truth, we typically recommend the default 28-day click and 1-day view. Facebook is great for generating direct response, but its ad formats are also a great platform to educate new users about your brand, product, or service. This is a good, conservative way to account for this total value of Facebook advertising when you’re comparing Facebook spending directly to other ad channels and their own associated attribution metrics.
When measuring mid- and bottom-funnel audiences like remarketing, we also recommend using a 28-day click model for performance evaluation. This user is familiar with your brand and your product, and in many cases this person is eligible for retargeting across a variety of platforms. While the view here can be valuable in reminding the user that they checked you out, it’s often a little too generous to give a Facebook ad view credit for the action.
For top-of-funnel efforts like prospecting (lookalikes, job titles, interests, etc.), we typically again recommend using the default 28-day click and 1-day view. In the case of an individual who has never heard of your brand, product, or service, a view on Facebook can be the catalyst they need to search the brand name or go direct-to-site. Personally, on what seems like a daily basis, I see an ad for a brand I’ve never heard of and write it down to do more research when I’m not on Facebook, like when I get off the bus to go home. Such is the power of the impression. We include the 1-day view to credit that impression and the non-zero impact of driving a user to our site. You can also extend the view credit to 7 or 28 days, but we typically look for compelling purchase path evidence before recommending longer view windows.
While there is no “correct” attribution model to use, the best option is the one that all parties agree to. It can be one of the most difficult choices to find consensus on, but I hope the factors outlined above help you and your teams reach one that’s well-considered. Reach out to our team for more guidance on picking your attribution model.