Attribution sounds like jargon until you watch it solve a concrete problem. So let us use the hardest concrete problem there is, because the hardest version makes the idea clearest: proving that a physical ad, which literally nobody clicked, actually caused a digital action. If you can follow the reasoning there, you will understand attribution better than most people who use the word professionally.
Start with the plainest possible definition. Attribution is the attempt to assign credit for an outcome to the thing that caused it. That is the whole idea. Something good happened, a sale, a signup, a visit, and you want to know what actually caused it, so you can do more of that thing. Simple enough. The trouble is that the easy version of attribution is not really attribution at all, and most beginners learn the easy version first and never find out it is broken.
The easy, broken version is called last-click. In a lot of digital tools, when someone converts, the system credits whatever ad or link it happened to see most recently before the conversion. It sounds reasonable until you think about it for one more second. The last thing someone saw is not necessarily the thing that caused them to act. Maybe they had already decided, and the last click was just the final step of a journey that something else set in motion. Last-click has no idea what would have happened otherwise. It is not measuring cause. It is labeling the last thing it saw and hoping that counts.
The honest version does something harder, and here is the whole trick in one idea: you have to imagine what would have happened without the ad, and give the ad credit only for the difference. That imagined version, the world where the ad never ran, is called the counterfactual. You cannot observe it directly, obviously, because the ad did run. So you approximate it with a comparison. You find a group of people, or a region, that was exposed to the ad, and another group that was similar in every way you can manage but was not exposed. Then you look at the difference in what they did. If the exposed group acted more, and the two groups were genuinely comparable, that difference is your evidence that the ad caused something.
Now watch why a billboard is the perfect teaching example. There is no click to fall back on. You cannot cheat by pointing at a last click, because there is no click at all. So you are forced into the honest method from the very start. To measure whether a billboard worked, you estimate who was likely exposed to it, often using anonymized location data that shows which phones passed through the area, and you compare their later behavior against a similar group who were not in that area. If the people who passed the billboard later visited the website or the store at a higher rate than the comparison group, you have real evidence the billboard did something. That is advertising attribution for physical media, and it lives or dies entirely on the quality of the comparison. Get the comparison right and you can credibly say a billboard moved behavior, even though no one ever tapped it.
Notice what just happened. The channel that seems impossible to measure, because it has no click, actually forces the most honest kind of measurement, because it has no click. The absence of the easy shortcut is what makes the method rigorous. Meanwhile the channel that seems easy to measure, the clickable digital ad, is the one where people most often fool themselves, because the click is right there tempting them to skip the hard question.
Once you have seen the idea work in the hardest case, it becomes easy to carry everywhere else, and it makes you much harder to fool. The next time someone shows you a marketing dashboard full of conversions and tells you a campaign worked, you will know the right question to ask, and it is a devastatingly simple one: compared to what? Compared to people who did not see it? How do you know those people were similar? Where is the counterfactual? Most confident-looking marketing numbers fall apart the moment you ask that question, because most of them never had a comparison in the first place.
So here is the one sentence to remember, the whole of attribution compressed into something you can actually use: cause needs a counterfactual, and everything that skips the counterfactual is a tally pretending to be a conclusion. A tally counts what happened. A conclusion tells you what your action changed. They are not the same thing, and learning to tell them apart is most of what separates real measurement from expensive guessing. The humble billboard, of all things, is one of the clearest places to learn the difference, precisely because it gives you no way to cheat.
Comments
Loading comments…