Data-driven Attribution

Data-driven attribution assigns conversion credit by comparing the paths of converting and non-converting users instead of applying a fixed positional rule.

The mechanism estimates each touchpoint's share from the difference in conversion probability between paths that include it and paths that do not. That method carries two serious limits. It needs a high volume of conversions and varied paths, and on thin data its output is unstable. And despite the name, it still observes correlation rather than causation: it runs no experiment and cannot say what happens if a channel is switched off. It is also opaque, since the shares cannot be reproduced by hand and shift as the data shifts, which makes month-over-month comparison of one channel's share risky.