If you’re a regular partaker of our bloggage, you will be used to us talking about econometrics. Yes, we do it. Yes, we love it. In this article, we thought we’d give everyone a break by talking about something other than specifically econometrics. But before anyone fears that you’re going to have to read my pontifications on renewable energy or the state of the world today, I will confine myself to marketing attribution. And econometrics will have, if not a starring role, something more than a walk-on cameo.
Life: There’s More to it Than Econometrics
The variety of approaches to marketing attribution
Did you know that econometrics is not the only approach to marketing attribution? There are, unsurprisingly, a good half dozen or so common approaches to marketing attribution, all of which try and size the contribution of your marketing efforts and put a value thereupon.
And all, of course, with their own strengths and weaknesses. Take television spot-matching, for instance. That offers super-granular (i.e. spot level) attributions of response (e.g. web visits, registrations, sign-ups) and so allows you to identify exactly what spots – what kind of spots – work best for you. Great for fine-tuning the TV mix. But as the allowed response window is typically 5-6 minutes from the time the spot airs, it’s only really giving you a partial picture of response and so, unsurprisingly, when we model the same thing using econometrics, we tend to see much bigger attributions and lower Costs per Acquisition (CPA).
Spot-matching is something we’ve written about before, read the blog. In this article, we compare three of the most common approaches to marketing attribution with econometrics and see how similar – or otherwise – the results are. I should say at the outset, this is not an exercise in determining The Best Approach to attribution: what’s best for you very much depends on your situation, the marketing activity you do and the questions you are asking.
Let’s start where everyone tends to start – Google Analytics.
Google Analytics (GA) vs. Econometrics
The theory behind Google Analytics
If you’ve thought at all about attribution, then you’ve undoubtedly thought about Google Analytics. Since the advent of the digital era, it has become the staple of brands’ attempts to measure their marketing. And it’s not hard to see why: as approaches go, it’s easy, cheap and intuitive. Plus you, Marketing Manager, don’t need to do anything. All it needs to work is a GA account and a small piece of code to be added to each page on your website, which sounds very much like a job for IT. One up and running it will tell you how users interact with your site (useful for optimising their journey), what hardware and browser they’re using (less useful) and, critically, how they arrived at your site.
From a marketing attribution point of view, traffic source is the good stuff. It will tell you how many sessions or users in a week, or similar, came to your site directly (by typing your URL) by searching, by clicking on a banner ad, a link in an email and so on. Talking of search, it can also tell you what sort of search your visitors performed, whether they searched for your brand name (a brand search) or for a generic term, such as “car insurance” (a generic search, predictably enough). Sweet stuff.
Limitations of Google Analytics
There are only three problems with it, unfortunately, all fairly fundamental.
The first is obvious. You’re tracking the source of activity to your website. All well and good if your website is the dominant means by which your customers interact with your brand. Less good, if you sell shampoo in supermarkets. I mean, people might come to your website to validate the greenness of your supply chain, but you’d probably rather they went and bought some in Tesco. And, therefore, you’d probably rather know what drove them to make their purchases, which is hardly the same thing. So, in effect, you’ve collected a lot of information about a small subset of your customers which is only tangentially related to the thing you really care about – why they bought your product.
Secondly, in our GA tracking we’re not able to view, nevermind measure, the effect of any offline activity to which our customers might have been exposed. Did they watch a TV ad? Don’t know. Did they walk past a particularly effective poster? Don’t know. Did they read about your brand in a magazine article? Don’t know. Some clients work around this by equating “direct traffic” to their website with the offline media they’ve done. But that’s really no better, as much of the direct traffic will be due to non-marketing drivers such as seasonality and market demand, and, conversely, some of the offline media effect will be picked up elsewhere, as we’ll see next.
The third is more subtle, but just as fundamental. What have we really tracked with our GA? The traffic source or, to be more precise, the referrer. In the case of a display banner that may not be so bad, but for something like brand search it’s bordering on egregious. What made you search for something in the first place? It wasn’t “brand search”, yet that’s what is getting all the credit for the visit here. This is so serious a problem it has its own name – the “last-click problem.”
Contrasting GA with Econometric Attribution
Unsurprisingly, when compared to econometric attribution, GA tends to overattribute credit to search channels and underattribute other channels. In the example below (taken from a real example, suitably anonymised), GA wants to give credit for 28% sales to Branded Search, more than twice the credit given by econometrics (13.2%). Conversely, for generic search, where there is typically more incremental effect, GA underestimates the contribution compared to econometrics.
The chart also illustrates the danger of relying on ‘Direct’ (which makes up the majority of Base in the chart) as the measure of offline media effect in GA. It is more than 20% points less than the actual Base contribution, as measured by econometrics. In short, relying on GA alone would lead us to misattribute almost every channel, leading to some wrong old decisions.
The Bottom Line
Quick, cheap and easy, you can see why GA attribution was so widely adopted in the early years of digital media – and why most clients have since had second thoughts about relying on it. It’s not without value, but you should never really look at it without performing a few mental calculations to adjust for its obvious biases.
Look out for our next instalment on marketing attribution alternatives, we’ll be looking at Spot-Matching. In the meantime, download our free econometric guide or drop us a line.