In October of 2012, Facebook made a claim that should be scary to any data-driven marketer: clicks hardly matter. Citing a study they conducted with Datalogix and research by Neilson, they argued that impressions were far more important than the click-through rate. With only a .07 percent correlation between high click-through rates and sales, and a striking conclusion that 99 percent of sales resulted from an impression, not a click, the numbers are startling.
Traditional marketers have known for some time that brand impressions influence sales, but this kind of thinking can make digital marketers uneasy. We're used to tracking actual visits, attributing them to traffic sources, and directly measuring where sales are coming from.
There's reason to be skeptical of these exact figures, but if impressions have anything remotely close to this level of influence, we couldn't call ourselves data-driven if we just dismissed the possibility. Last week, I argued that it's more important to track Facebook referrals and return visitors than "likes." Today, I'm arguing that it's just as important to attribute sales to impressions, regardless of traffic.
Where is Your Direct Traffic Coming From?
Direct traffic is the red headed stepchild of Analytics. We like seeing it, but we tend to neglect it, and we don't give it the credit it deserves. This is, in large part, because we don't know where the direct traffic is coming from. Without measurement to attribute this traffic, we don't know where to invest our efforts.
It is possible to estimate how much of our direct traffic results from Facebook impressions if we have access to statistical software. Let me show you how I used Minitab to estimate the influence Facebook impressions had on my science blog.
First, I exported data from Facebook Insights to get the number of impressions. Then I exported my direct traffic data from Google Analytics. I copied the daily, weekly, and 28 day impression columns from Insights over to Minitab. Then I copied the direct traffic data over, making sure that the dates lined up.
Next, I went to the menu and clicked on Stat: Regression: General Regression.
I set Direct Traffic as the response and Daily as the model. In other words, I asked Minitab to create a mathematical model that would predict my direct traffic based on my daily Facebook impressions. (I actually tested this with the weekly and 28 day impressions, but daily was the best fit).
After hitting OK, I got this:
This equation is a predictor of how much direct traffic I can expect to get in relationship to my impressions on Facebook. That coefficient 0.000554703 means that I can expect to get about 1 visitor for every 2,000 impressions. I was running ads at the time, so this tiny figure doesn't surprise me. (If I wanted to be more rigorous, I could have looked at organic impressions vs paid impressions).
What's more interesting is this:
In particular, under that column labeled "P," in the row labeled "Daily." That value, 0.022, is called the P-value. You'll want to pay attention to it if you're doing this for yourself. It's the probability that we would get results like this if Facebook impressions had no influence on Direct traffic. In other words, if Facebook impressions legitimately had no influence on my Direct Traffic, there's only a 2 percent probability we'd get results like this. That's actually significant enough for most scientific studies, which look for P-values less than 5 percent.
Now, I wouldn't be a very good statistics major if I didn't point out that correlation isn't causation. We can't discount the possibility that the direct traffic was leading to Facebook visits, which increased impressions. Since the vast majority of these impressions are from ads, though, I find this unlikely.
So yes, it is possible to have some degree of attribution, even when you're dealing with direct traffic. We're dealing with estimates, not hard numbers, but the results are enlightening. The estimated number of direct traffic visits I got from impressions was comparable to the number of click-through visits I got from ads, so it's certainly worth paying attention to.
While I've gone through this process for direct traffic, there's no reason you can't go through the same process for sales. Admittedly, this is fairly rudimentary analysis, and it's worth having a statistician on board if you want better estimates. (In particular, the assumption of same day purchases or site visits is a hard sell.) But some data is better than no data, and you'll only err on the conservative side with this technique.
Do you estimate where your direct traffic is coming from? Do we as digital marketers underestimate the influence of impressions?
Image credit: Kevin Dooley