Attribution You Can Defend in the Boardroom

The precision was always borrowed

For a decade marketers presented last click attribution to boards as though it were accounting. It never was. It was a convention: a rule for assigning credit that happened to be cheap to compute and hard to argue with in the room, because the number carried two decimal places and the alternative was a hunch.

The loss of third party identifiers removed the convention without removing the expectation. Directors still want to know what the spend did. The honest reconstruction is not a better tracking pixel. It is a different kind of claim, made with different evidence and defended on different grounds.

Three sources, none of them sufficient alone

Modern marketing measurement rests on a triangle, and the discipline lies in reconciling the three rather than picking a favourite.

• Experiments: geographic splits, holdouts and switchbacks that establish cause directly, at the cost of time and reach
• Models: media mix modelling that attributes outcomes across channels, useful for allocation, weak on short term detail
• Observed data: platform reporting and first party signals, precise about behaviour they can see, blind to everything else

When the three agree, act with confidence. When they disagree, the disagreement is the finding, and chasing it will usually teach you more than another dashboard would.

Incrementality is the question being asked

When a director asks whether the brand campaign worked, they are not asking which touchpoint deserves the credit. They are asking what would have happened if it had not run. That is an incrementality question, and it can only be answered by comparison, never by attribution logic applied to the people who happened to convert.

The practical form is a holdout. Withhold spend from a region, a segment or a period, then measure the difference. Australian teams have an advantage that global teams envy here: state and metropolitan markets are distinct enough to form usable comparison groups, and the population is concentrated enough that a geographic test reaches meaningful volume quickly.

The objection is always the same, that a holdout means giving up sales. It does, and that is the price of knowing. A team that has never once suppressed spend to see what happens is not measuring. It is reporting.

Say the uncertainty out loud

The instinct when presenting a modelled result is to strip out the range, because ranges look like weakness. That is backwards. A single number invites the board to treat it as fact, and the first time reality diverges from it, the whole measurement program loses its standing.

Present the estimate, the range, and the assumption most likely to be wrong. Say which way the answer moves if that assumption fails. This is how every other quantitative function in the business already speaks, and marketing is the last one still performing a certainty it does not have.

A range you can defend is worth more to a board than a point estimate you will spend the next four quarters quietly walking back.

Rebuild the first party base

None of this works without owned data that means something, and that is a smaller task than most vendor pitches suggest. A reliable identifier for the customers you already have. Consistent event definitions across the site and the shop. A record of what was spent where and when, kept so that a person can query it without raising a project.

The unglamorous half of measurement is the spend log. Most Australian teams cannot reconstruct, a year later, exactly what ran in which market with which creative. No model rescues that, and no consultant can build it back after the fact.

What to bring to the board

One page. The decision you want, the evidence behind it, your confidence in that evidence, and the test that would change your mind. Directors handle uncertainty in every other paper they read. What they cannot use is a marketing function speaking a private dialect and producing numbers nobody else is able to interrogate.