What is marketing attribution? (and why you might want to forget all about it)

Let me start by telling you a story:

Many, many moons ago I ran email marketing for a large B2C organization. Excited by the new-found ability to attribute revenue to email clicks, I created and ran a report detailing the total amount of business my emails were delivering.

It was quite a small number, but it was probably enough to claim ROI (newflash: most of marketing involves messing around with numbers until you can ‘claim ROI’).

Only later did a particularly perceptive individual (who may or may not have been myself) point out the obvious, namely:

• I’m sending email to an audience that is literally defined as my most valuable, most regular customers, and

• some of them are clicking on the email and spending, but

• a sizable proportion of them probably would have spent this weekend anyway, so

• how do I know I’ve made any difference to the business at all?

I’ll discuss the answer to that question in more detail below, but for now let’s just say this was my first realisation that….

Attribution is complicated

What I learned that day was the importance of what Edward Tufte has called “the question at the heart of quantitative thinking”:

“Compared to what?”

Or to put that a little more clearly, REAL attribution, and in fact the accurate measurement of ANY marketing activity, means being able to confidently show the difference between doing something and not doing something.

Many years have passed since I realised this. But I still see people make this mistake all the time. I still find myself having this conversation, all the time:

• Anonymous marketer: “We sent voucher codes in emails to try and bring back lapsed customers, and 57 customers came back giving us an ROI of 528%”

• Me: “Do you know how many of that audience would have come back anyway?”

• Anonymous marketer: “No”

• Me: Weeps silently, prays to St Bernardino*

And before you ask, yes, this was something I learned the hard way too.

The thing is, the particular challenges in the stories above are actually relatively straightforward to solve. All you need is a control group. But when we start attempting attribution for acquisition, things get really confusing. Let’s talk about it.

Attribution for acquisition is really complicated

Ever since marketers have been able to measure things they have tended to make the fatal assumption that everything important can be measured.

This is completely untrue and the first step in my recovery program is to stop believing it.

Even if human beings were perfectly rational actors (spoiler: they are not), it is completely impossible to pick apart the individual elements that contribute to a multi-channel marketing campaign and say with any great certainty which elements deliver ROI and which do not.

There are lots of reasons for this, but some of them include:

• The absence of control groups. Almost by definition these cannot exist for acquisition campaigns, because there is no way of randomly showing ads to one group rather than another (which is not the same as randomly showing alternative creative within a campaign, which is OK).

• The existence of an incredibly complex web of potential touches and interactions with the consumer, some of which are trackable, and some of which are not.

• The virtual impossibility of truly measuring the impact of any one ‘touch’ within a campaign

• The inability to account for word-of-mouth, or personal responses to the brand that are almost impossible to quantify.

• The lack of any long-term analysis. Results we see after a week may be due to a shock factor that actually puts people off our brand and product in the long-run.

I could keep adding to that list all day, but you would soon be even more bored than you are already.

Of course, none of this means that measurement should never be attempted. But it does mean that we should accept that perfect knowledge isn’t a thing, and act accordingly.

Aside from anything else, “you can’t step into the same river twice” as Heraclitus tells us. The whole point of attribution is to learn from yesterday in order to be more effective tomorrow, but tomorrow can never be truly the same as yesterday. Things will always have changed, and we need to be aware of that.

Before giving you a few pointers around attribution, let me just make sure one thing is clearly understood. I am not against data. I love data, and you should too. But I am in favour of understanding the limits of data. Which leads me to...

Five ways to get good (or at least better) at attribution

So with all that said, how about some positive advice about this subject?

Here’s five ways to think about attribution that might help:

1. Match your approach to your business. If you sell widgets to consumers and your sales cycle from awareness to purchase is short, there is absolutely nothing wrong with using a conventional attribution approach. Another way to think about this would be, if your first touch and last touch are kind of the same, attribution works, and you should feel quite confident about using it. If, on the other hand, you have a long sales cycle involving lots of touches and interactions, think about not using attribution at all. By forcing every deal to have a ‘source’ you are flattening a huge amount of complexity into a very simplistic view of the world. Which is dangerous.
   

2. Try to learn in a structured way. Learning that a specific ad was effective in the past tells us nothing, other than that specific ad might be effective in the future. But if we classify ads into types, we can start to reach tentative conclusions about what type of creative is more effective than others. This is useful information (allowing for all the caveats everywhere else in this piece) - much more useful that ‘random ad A was more effective than random ad B’.
   

3. Be careful about attribution around ‘universal’ activities. A lot of multi-touch attribution shows you in detail all the various things your customers saw or responded to on their ‘journey’ to a purchase. But some of these things, almost everyone sees. Re-targeted display ads based on site visits is a good example. So are ‘nurture’ emails sent after an initial interaction. Don’t assume they are fantastically powerful just because your analysis tells you they are involved in every deal you do. There would be a problem if they weren’t.
   

4. Focus on key moments. I may sound like a luddite when I say this, but there’s a hell of a lot to be said for first touch attribution. The change of state from “I have never interacted with this company” to “I have registered for a webinar” (just to give one example) is huge. Of course there may be things this person has seen before, and that’s all good, but one specific thing, in one specific channel, got them over the line. Identify the changes in state that really matter to you, and what makes them happen, rather than looking back after the close at a sea of activities all claiming to be part of that success, which brings me to...
   

5. Never believe third parties. Speaks for itself. I have no interest in a third party telling me that a new customer saw a display ad on their network 3 months ago. There is way too much sharp practice in the digital advertising industry to take that kind of thing on trust anyway.

There is so much more that could be said on this subject, and I suspect I’ll pick up something specific from here next week, but that’s probably enough to be getting on with - I have work to do after all.

*The patron saint of marketers, remember?