Data-Driven Attribution (DDA) is a class of attribution models that assigns conversion credit to each touchpoint on a converting path using a model trained on observed paths, rather than a fixed rule like last-click or linear. What makes it data-driven is that the credit weights are fit to a brand’s (or platform’s) own conversion paths rather than declared up front. Most operators meet the term when Google Analytics 4 or Google Ads surfaces it as the default attribution model — Google moved both products to DDA defaults in 2023.
The canonical textbook reference for what such a model is doing is a Shapley-value decomposition from cooperative game theory: each touchpoint’s credit is its average marginal contribution to conversion probability across all possible orderings of the path. Google’s production model is machine-learning-based and the exact variant is not publicly documented, so treat Shapley as the conceptual frame rather than a description of the shipped system.
DDA sits inside the multi-touch attribution category — alongside linear, time-decay, position-based, and U-shaped — and is distinct from marketing-mix modeling, which is top-down on aggregate spend and geo/time data with no user-level paths. The category boundary matters because operators frequently mix the two when comparing tools.
What DDA can do: redistribute credit across a converting path more faithfully than last-click attribution, which systematically under-credits upper-funnel impressions and early clicks by rewarding only the terminal event. What it cannot do: recover paths the platform never observed (the iOS, ATT, and consent-denied gap reduces input quality, and DDA inherits it), answer incrementality questions (DDA distributes observed credit; it does not tell you which conversions would have happened anyway), or be audited by the brand the way an in-house attribution layer can be. The platform exposes outputs, not the model.