What is fractional attribution?

The idea

Single-touch attribution answers a question nobody asked — which one thing gets the credit — when the honest answer is usually that several things contributed.

Fractional attribution assigns each touchpoint a share. A conversion becomes 0.4 to paid search, 0.3 to a video impression, 0.3 to a retargeting click, rather than 1.0 to whichever happened last. Reporting becomes decimal, which is unfamiliar and considerably more truthful.

How the fractions get decided

• By rule. Linear, time decay or position-based weights, applied uniformly. Transparent and arbitrary.
• By model. Weights learned from comparing converting and non-converting paths. More defensible, and dependent on seeing complete paths.
• By experiment. Fractions calibrated against measured lift. The most credible and by far the most expensive.

How to read fractional numbers

Two failure modes are common enough to name.

Treating fractions as precision. A 0.37 share is the output of an assumption, not a measurement. Reporting it to two decimal places implies a confidence the method does not support.

Optimising a channel to its fraction. A channel with a small share of many conversions may be doing the most valuable work in the account. Cutting it because its number is small is exactly the mistake fractional attribution was invented to prevent, arriving through a different door.

The useful test is the same as for any attribution output: does a holdout agree? Where fractional attribution and incrementality disagree, the holdout is the evidence and the fractions are the hypothesis.