Why Multi-Touch Attribution Is Mostly Theatre

Multi-touch attribution promises to tell you exactly which touchpoints produced a sale and how much credit each deserves.

It cannot, and the reason is not that the tools are immature. It is that the question does not have a measurable answer.

What the models actually do

Every attribution model is a rule for splitting credit, chosen by a person.

First touch gives everything to the first interaction. Last touch gives everything to the last. Linear splits it evenly. Time decay weights the recent ones. Position-based weights the first and last.

None of these is derived from evidence about how the customer decided. They are assumptions with a dashboard attached, and they produce different answers from the same data.

If changing the model changes which channel looks best, the model is deciding your budget rather than the data.

What is genuinely unmeasurable

The touchpoints nobody sees. A recommendation from a friend. A conversation at an event. Something seen months ago and half-remembered. None of these appear in any attribution system and they routinely matter more than the ones that do.

Cross-device journeys. Phone, laptop, work computer. Stitched together imperfectly at best.

Blocked and consented-away tracking. A meaningful share of your audience simply is not measured.

View-through influence. Someone sees an ad, does not click, and searches your name two weeks later. The search gets the credit. The ad did the work.

Why the platforms disagree with each other

Each ad platform reports conversions it believes it influenced, using its own window and its own rules, and each one attributes to itself.

Add up what every platform claims and you will exceed your actual sales, often substantially. Nobody is lying. They are each measuring their own contribution in isolation.

Expecting them to reconcile is expecting three people to divide credit for a decision none of them witnessed.

What to do instead

Measure incrementality, not attribution. Turn a channel off, or hold out a region, and see what happens to total revenue. That is a real experiment and it answers the only question that matters: what changes when I stop spending here.

This is more work and less granular than a dashboard, and it is the only method that produces a defensible answer.

Ask the customer. A single question on the enquiry form, "how did you hear about us," in a free text field. The answers are imperfect and they capture the touchpoints no system sees.

Watch the total, not the split. Total spend against total revenue over a period. Blunt, unfashionable, and impossible to game.

Use platform attribution for optimisation only. The algorithms need conversion feedback to work. Give it to them. Just do not use it to settle budget arguments between channels.

The practical position

Attribution models are useful for spotting a trend in one channel over time, because the bias is at least consistent.

They are not useful for deciding that one channel deserves more budget than another, because that comparison is exactly where the bias lives.

Run the experiment instead. It costs a fortnight of held-out spend and it tells you more than a year of dashboard analysis.

Where this leaves brand

The channels hardest to attribute are usually the ones building the demand the measurable channels then capture.

Organic content, in particular, produces the searches that paid search converts and the familiarity that makes an ad work. Judged on last-click it looks weak. Turned off, everything else gets more expensive.

That relationship is why our own model runs content organically first and puts spend behind what has already proven it holds attention, rather than treating the two as separate lines on a budget.