What is Data-Driven Attribution?

Data-driven attribution (DDA) is a model that splits conversion credit across touchpoints by estimated contribution to conversion probability, not by a fixed rule. The algorithm compares paths that reached the goal with paths that did not, and gives more weight to interactions without which conversions happen less often.

The English query is the demand: people search data-driven attribution and data-driven attribution GA4, not a Russian calque. Yandex suggest is thin on the RU phrase — the H1 keeps the English term.

In GA4, data-driven is the default reporting model. Since late 2023, property attribution settings offer DDA and last click (paid and organic last click; separately, Google paid channels last click). That is not a CPA-network payout rule and not a replacement for a tracker.

How DDA differs from last-click and from rule-based MTA

Last-click attribution gives 100% to the final click (in GA4, usually the last non-direct touch). Easy to compute, easy to dispute, and how most affiliate networks pay.

Multi-touch attribution is the family of models that split credit. DDA is the ML member of that family. Next to it sit rules that GA4 no longer offers as cross-channel models but that still appear in other tools and older reports. They stay sections here, not new slugs.

Linear

Linear gives every touch on the path the same share. Four clicks before a purchase → 25% each. The model does not argue which channel “matters more,” and therefore overweights the middle of the path: a mid-funnel display or a repeat push gets as much as closing search. Useful as a teaching baseline and a conservative check, not as the only bidding KPI.

Time-decay

Time-decay increases weight toward the conversion: yesterday’s touch counts more than one from three weeks ago. A half-life parameter is typical. The model sits closer to last-click than linear does, but it does not zero the top of the funnel. On a same-day lead cycle, time-decay almost equals last-click; on a long cycle (finance, subscription) it still leaves a share for early warming.

Related, and still not separate pages: position-based / U-shaped (often 40% first, 40% last, 20% middle) and first-click (100% to the first touch). GA4 removed those rules in 2023 together with linear and time-decay. Google Ads may still show its own reports in places, but cross-channel analytics default is DDA.

How data-driven is computed (idea, not an Ads formula)

The platform looks at channel sequences. If the path “paid search → email → direct” converts clearly more often than the same path without paid search, search takes a share even if last-click would have given everything to email or direct. If retargeting is almost always last on an already-warm user, DDA will cut its weight versus last-click — a common report conflict.

The estimate is specific to the account, the conversion type, and the data window. Two advertisers with different channel mixes will see different weights on the same source names. With too few key events the model effectively falls back to last-click even when the UI still says “data-driven.”

Where DDA lives and where it does not

GA4 — reporting default; fractional conversions in Advertising (0.3 + 0.7) are expected, not a bug.
Google Ads — DDA by default on eligible conversion actions; imports from GA4 also follow data-driven logic when the model is trained.
Meta and other ad accounts — their own modeled / data-driven reports; you cannot add them arithmetically to GA4.
Affiliate networks and the tracker — payout is almost always last-click on a click ID. DDA does not rewrite the postback.

Yandex Metrica uses its own models (last / last significant / first / automatic). Metrica’s “automatic attribution” is not GA4 DDA.

Limits

DDA is a black box: you cannot write the weights as a slide formula. Privacy (ITP, ATT, consent) punches holes in the path, and the model fills gaps. On SKAdNetwork there is no user-level DDA — only aggregates. Reconciling the ad account, GA4, and the tracker without naming the model is pointless: those are different definitions of “whose conversion,” not lost data. For a buyer, DDA is useful so last-click retargeting does not starve prospecting; for offer accounting it is not the source of truth.