Attribution Model Calculator

How the figure is built

An attribution model allocates credit for a conversion across the touchpoints that preceded it. Last click gives all of it to the final interaction; first click to the initial one; linear splits it evenly; time decay weights recent touches more heavily.

None is correct. Each is a rule for dividing something that cannot be observed, and the model chosen determines which channels appear to work.

Credit assigned = conversion value × the model's weight for that touchpoint.

An example

A customer sees a social ad, reads a blog post from search, receives an email, then converts through a branded search click. One $58 order, four touchpoints.

Last click gives branded search all $58. First click gives social all of it. Linear gives each $14.50. Time decay might give $29 to the branded click and the rest to the others.

Under last click, social appears worthless and branded search appears excellent. Under first click the reverse. The customer journey was identical.

A business reallocating budget on last-click data will steadily defund the channels that create demand in favour of the one that harvests it, which works until there is no demand left to harvest.

Where the figure deceives

Every model shares the same fundamental flaw: it distributes credit among tracked digital touchpoints and ignores everything untracked: word of mouth, offline exposure, brand familiarity built over years.

Signal loss has made this worse. Cross-device journeys, blocked cookies and privacy restrictions mean a growing share of the path is simply invisible, and models fill the gap with assumptions.

Acting on it

Pick one model, understand its bias, and use it consistently for within-channel optimisation rather than for budget allocation between channels.

Then use holdout tests for the allocation decisions. Only an experiment measures whether revenue exists because of a channel, and no attribution model can substitute for it.

Why the model choice is a strategic decision

Last-click attribution systematically favours bottom-of-funnel activity, which produces a self-reinforcing loop: budget moves to retargeting and branded search, demand-creating channels are defunded, the pipeline of new demand shrinks, and the harvesting channels eventually run dry.

The decline is slow and the metrics look excellent throughout, which is what makes it dangerous. Reported ROAS improves the whole way down.

That is the strongest practical argument for measuring incrementality periodically even though it costs revenue during the test. It is the only measurement that would catch this pattern, and businesses that discover it usually do so after a year of unexplained decline in a channel that was never being credited.

Comparing the same period under two different models is a cheap way to see how much the choice is driving the conclusion, and most analytics platforms allow it without any additional setup.

Where two models disagree sharply about a channel, that channel is the one whose value should be tested experimentally rather than argued about.

Documenting which model is in use, and when it last changed, prevents the recurring confusion where a shift in reported channel performance turns out to be a settings change.