Myth: Google's data-driven attribution tells you what's incremental

Myth: Google's data-driven attribution tells you what's incremental

The question: when Google Ads switched everyone to data-driven attribution (DDA), did practitioners gain a causal read on which touchpoints actually drove conversions?

What the methodology says: DDA uses a Shapley-value approach — a cooperative game-theory method (originally from economics) that distributes credit by comparing converting and non-converting paths. It is genuinely better than last-click at describing observed paths. But it is built entirely on observational data: it sees correlations between exposure and conversion, not counterfactuals.

The nuance: Shapley fairly allocates observed credit, but it cannot answer 'what would have happened without this channel.' If brand search appears on most winning paths, DDA rewards brand search heavily — even when those users would have converted regardless. Google's own documentation calls DDA a credit-distribution model, not an incrementality measurement. Meta-analyses of geo-lift experiments routinely find DDA over-credits cheap, late-funnel, high-intent channels by 20-50% versus randomized holdouts.

This is the classic correlation-causation gap: presence on a path is correlated with conversion; it does not prove the touch caused it.

Bottom line for practitioners: treat DDA as a sophisticated descriptive lens for reallocating within already-validated channels. For the decision that matters — should this channel exist at all — you need an experiment (geo holdout, ghost ads, randomized PSA control), not a smarter way of dividing post-hoc credit.