Half my advertising is wasted, Wanamaker is supposed to have said; the trouble is I do not know which half. The remark survives because everyone assumes the problem was ignorance. It was not. It was that the question has no answer of the kind he wanted.
A century later marketing departments hold more behavioural data than any discipline outside epidemiology, and they still cannot say which half. What changed is that they can now produce a number, and the number has the shape of an answer.
The last click was never the cause
Attribution models assign credit for a purchase to the touchpoints that preceded it. Every such model rests on an assumption that is false and known to be false: that the touchpoints observed are the ones that mattered.
The customer who searched the brand name and clicked the ad had already decided. The search was not persuasion, it was navigation — and last-click attribution hands that conversion to the paid channel that intercepted it, while the conversation with a colleague that actually caused the decision leaves no record at all. Multi-touch models spread the credit more widely but do not fix the defect, because they still distribute credit only among the events the measurement system can see.
What the system cannot see is not a small residue. It is most of the mechanism: a recommendation, a shelf noticed in passing, an article read two years ago, a reputation absorbed without any specific encounter at all. Marketing works largely through memory, and memory does not emit events.
Why more data made this worse
One would expect richer data to narrow the gap. It widened it, for a reason that is easy to state and hard to accept: the measurable share of the mechanism grew faster than the unmeasurable share, so the measurable share looked increasingly like the whole.
Budgets follow measurement. A channel that reports a return gets funded; a channel whose contribution is real but unattributable does not. Over several planning cycles this produces a portfolio optimised for legibility rather than for effect — heavy on the activities that capture existing demand, light on the activities that create it. The numbers improve throughout. The business does not.
This is not a hypothetical failure mode. It is the ordinary history of the last fifteen years of digital marketing budgets.
The question worth asking
The productive move is to stop asking which touchpoint caused this sale, a question the data cannot answer, and start asking whether the spend caused any sales at all — a question it can.
That question is answered by comparison, not by attribution. Hold out a region and spend nothing there. Stagger a campaign launch across markets. Switch a channel off for a fortnight and watch what happens to markets that never saw it. These designs are unfashionable because they are slow, because they cost something in foregone activity, and because they occasionally deliver the answer that a well-liked channel does nothing. They are also the only designs that establish causation rather than coincidence.
Accountability is not the same as attribution
Underneath the technical argument is a political one. Marketing adopted attribution not primarily because it explained anything, but because it produced a defensible number in a room where finance was asking for one. That was a rational response to a real pressure, and it worked: it bought the function a seat.
The price was that the seat now depends on a number the function knows to be unreliable, which is a weak position to argue from when the budget is questioned. Accountability means being answerable for a result. Attribution means having a story about which click preceded it. They have been treated as the same thing for so long that the substitution is invisible.
A marketing function that can demonstrate, by experiment, that its spending produces incremental demand is in a far stronger position than one that can produce a dashboard. The dashboard is more comfortable. It is also the thing that will not survive contact with a sceptical chief financial officer who asks what would have happened if you had spent nothing.