Multi-Touch Attribution Calculator

The calculation itself

Multi-touch attribution distributes conversion credit across several touchpoints using a weighting rule, rather than assigning it entirely to one.

Position-based models, commonly 40% to the first touch, 40% to the last and 20% split across the middle. Are the usual compromise between simplicity and fairness.

Data-driven models derive the weights from observed conversion patterns rather than assigning them, which is better in principle and requires substantial volume.

Running the numbers

The same four-touchpoint journey under a 40/20/40 model: social gets $23.20, blog and email share $11.60, branded search gets $23.20.

Compared against last click, social has gone from $0 to $23.20 of attributed value and branded search from $58 to $23.20.

Across 1,000 orders that reallocates $23,200 of attributed revenue between channels, which at typical ROAS thresholds changes whether several campaigns look viable.

No underlying performance changed. The reallocation is entirely a consequence of the weighting rule chosen.

What gets missed

Multi-touch models look more sophisticated than last click and share its core limitation: they distribute credit among observable touchpoints only. A more elaborate rule applied to incomplete data is still applied to incomplete data.

Data-driven models also require conversion volume that most small businesses do not have, and below that threshold they fit noise rather than pattern.

What to do next

Treat multi-touch as a better lens rather than an answer. It corrects the worst bias of last click and does not measure causation.

Then check the model's conclusions against a holdout at least annually. Where the two disagree, the experiment is right and the model is a description of correlations.

What has replaced attribution at larger scale

Businesses with sufficient scale have moved toward marketing mix modelling, a statistical approach that relates aggregate spend across channels to aggregate outcomes, without needing to track individuals at all.

It is immune to signal loss because it never observes a customer journey, and it captures offline and brand effects that touch-based models cannot see. It requires several years of data and produces slower, coarser answers.

For a smaller business the practical middle ground is a simple version of the same logic: track blended efficiency, run periodic holdouts, and use attribution only for deciding what to change within a channel. That combination is more robust than any attribution model and considerably cheaper than a mix model.

Data volume determines whether a data-driven model is meaningful, and most platforms specify a conversion threshold below which they will not run one. Below it, a rules-based model is the honest choice.

Applying a sophisticated model to thin data produces confident-looking output derived from noise, which is worse than a simple model whose limitations are obvious.

Server-side tracking recovers some of the signal lost to browser restrictions, which improves the input data rather than the model and is usually the higher-return investment of the two.

Consent rates affect how much of the journey is visible at all, so a business in a strict privacy regime is modelling a smaller sample than one elsewhere and should hold its conclusions more loosely.