Last-Click Attribution Is Not a Neutral Default: A Short History of a Silent Bias

What last-click attribution actually is

Attribution is the problem of dividing the credit for one outcome among the several marketing touches that preceded it. A buyer might see a display ad, read a review, click a branded search result a week later, and finally arrive through a retargeting ad before booking. Attribution decides how much of that booking each touch earned. It is, unavoidably, a modeling choice, because the counterfactual (what the buyer would have done without any given touch) is never directly observed.

Last-click resolves that choice in the bluntest possible way: it assigns one hundred percent of the credit to the final touch and zero to all the others. Its appeal is that it requires no theory and no inference. The last click is a recorded fact, so the number is unambiguous, reproducible, and cheap to compute. Every other model, first-click, linear, time-decay, position-based, and the data-driven models that came later, is an attempt to spread the credit more realistically, and every one of them requires an assumption that last-click quietly avoids making.

That is the first thing to understand about last-click: it is not the absence of a model. It is a model with a strong and rarely stated assumption baked in, namely that only the final touch mattered. The apparent neutrality is the illusion.

How last-click became the industry default

A default is a powerful thing. In any system with millions of users, the setting that ships switched on becomes the near-universal behavior, because the overwhelming majority of people never change it. The history of attribution is, to a large degree, the history of what the dominant measurement tools chose as their out-of-the-box setting.

For most of the era in which digital marketing professionalized, the standard web-analytics reports credited conversions to the last campaign source a user arrived from, with a last-non-direct-click convention as the working default. Marketers did not sit down, weigh the models, and elect last-click on the merits. They opened the reports they already had, and last-click was the number the reports showed. The convention propagated through dashboards, agency templates, platform optimization, and quarterly reviews until it was simply what "the results" meant.

This is the argument at the center of this piece, and it should be read as an interpretation of the pattern rather than a measured finding: last-click achieved its silent dominance because it was the default of the analytics platforms most businesses used, not because anyone had shown it to be accurate. The corroborating evidence is that when the platforms changed their defaults, the field moved with them. Google Analytics 4 retired last-click as its standard model in favor of data-driven attribution, and the practitioner conversation shifted accordingly, which is exactly what you would expect if the default, not a proof of accuracy, had been carrying the convention all along.

What last-click systematically miscredits

A default would be harmless if it were unbiased. Last-click is not. Its error is not random noise that averages out; it is directional, and it always points the same way.

Because it rewards only the final touch, last-click over-credits the channels that tend to appear at the very end of a buyer's path and under-credits the ones that set that path in motion in the first place. Branded search and retargeting are the classic beneficiaries: they intercept people who have already decided to buy, so they are present at the moment of conversion and collect the credit for it, whether or not they caused anything. Upper-funnel work, the review that built trust, the content that created awareness, the brand impression that put the business on the buyer's shortlist, appears weeks earlier and never touches the final click, so under last-click it looks worthless.

The operational consequence is a slow, self-reinforcing misallocation. Budget flows toward the channels that are best at being last, which are frequently the channels best at claiming credit for demand someone else created. The genuinely demand-creating work looks inefficient by the same measure, so it gets cut, which weakens future demand, which the last-click report cannot see. This is the precise mechanism behind the owner-operator complaint that ad spend feels wasted: the measurement is steering the money by a number that is structurally blind to half of what the money did.

The evidence that observational attribution is biased

The strongest case against last-click, and against observational attribution generally, does not come from theory. It comes from a study that put the observational method next to experimental ground truth and measured the gap.

A very old problem wearing a new interface

It is tempting to treat attribution bias as a digital-age artifact, but the discipline has been circling this exact anxiety since long before the click existed. The most famous line in all of marketing, that half of advertising spend is wasted and no one knows which half, is usually attributed to John Wanamaker, and it names the problem perfectly.

The instructive detail is that the quote itself is unverified. The earliest documented trace is a secondhand 1919 account, and the same sentiment has been attributed to William Lever, William Wrigley, and others, so the industry's founding parable about the difficulty of measurement is, fittingly, a measurement it cannot substantiate. The lesson is not cynicism. It is that a convenient, quotable, computable answer, whether a folk quotation or a last-click dashboard, tends to get adopted and repeated for decades before anyone checks whether it is true. Last-click is the modern version of that pattern, and it deserves the same skepticism.

What is replacing last-click: experiments and mixed models

The correction to observational attribution is not a better observational model. It is a different kind of evidence: experiments that create a control group, and aggregate models that never needed to follow individuals at all.

Why this matters more for a small business, not less

There is a real caveat that any responsible reading has to state plainly. The rigorous alternatives to last-click, randomized experiments and firm-level mix models, were built on data volumes that a single-location business will realistically never reach on its own. A med-spa or a home-services contractor does not generate the traffic to power a clean geo-lift every month, and its spend is too small to fit a stable mix model from its own history alone. The scale floor is real, and ignoring it would misrepresent what these methods can do for a business this size.

The resolution is not a shortcut and it is not a smarter attribution setting. It is refusing to treat any single number as truth without a baseline behind it, reporting a blended efficiency figure that no individual channel can inflate, and, where accurate measurement requires more statistical power than one business owns, pooling evidence across many similar businesses to recover it. That is the corrective tradition last-click sits outside of, and it is the one worth continuing: measure the counterfactual where you can, disclose the method where you cannot, and never let a convenient default stand in for a proven result.