The essentials
- Definition: data-driven attribution is a model that splits the value of a conversion between the touchpoints in the journey according to their real contribution, estimated statistically by comparing the journeys that convert with those that do not.
- The difference with last click: last click gives everything to the final channel (often brand search or remarketing); data-driven also credits the channels that started or fed the journey, such as social advertising.
- Where: the default model in GA4 and Google Ads since 2023; Meta and LinkedIn have their own models and windows.
- Limits: it only sees the journeys that are measured (consent, devices, offline), it varies from one platform to another, and it is not proof of incrementality.
When a customer buys after seeing an Instagram ad, clicking a Google ad and reading an e-mail, which channel “won” the sale? The answer depends on the attribution model. Last click gives everything to the final channel; the data-driven model splits the value according to the estimated contribution of each touchpoint. For SMA, often early in the journey, this choice changes how return on investment reads. This article defines the model, explains how it works, what it changes, its limits and how to use it. The overview of attribution in GA4 is in attribution in GA4.
The attribution models compared
| Model | Split | Effect on SMA | Availability in 2026 |
|---|---|---|---|
| Last click | 100% to the last channel clicked before the conversion | Underestimates social advertising, which often starts the journey | GA4 (option), Google Ads (option), most tools |
| First click | 100% to the first channel | Overestimates social and Display | Removed from GA4 and Google Ads in 2023 |
| Linear, time decay, position-based | Split by a fixed rule | Arbitrary | Removed from GA4 and Google Ads in 2023 |
| Data-driven | According to the estimated contribution of each touchpoint, through statistical learning | Credits social for its real share in the journeys that convert | The default in GA4 and Google Ads |
| Social platform models | Meta: last touch with 7-day click / 1-day view windows; LinkedIn: last touch 30-day click / 7-day view | Each platform claims every conversion it touched | In the ads managers |
How the data-driven model works
- Collecting the journeys: for each conversion, the sequence of measured touchpoints (channel, campaign, order, delay), and the same for the journeys that did not convert.
- The comparison: the model estimates, for each channel, how much the probability of conversion rises when it is present in the journey, across comparable journeys (an approach inspired by Shapley values).
- The split: the value of each conversion is distributed between its touchpoints in proportion to that contribution; one channel may receive 0.3 of a conversion, another 0.7.
- Updating: the model relearns continuously; it needs a minimum volume (GA4: a few hundred conversions per month; below that, it falls back on rules).
The result: attributed conversions become decimal numbers, and the total per channel no longer matches last click. This is normal and expected.
What it changes for social advertising
- SMA regains credit: a Meta discovery campaign that precedes a Google search and a purchase receives a share of the sale; on last click, it received none.
- Remarketing and brand search lose some: they close journeys that others opened.
- Cost per conversion and ROAS by channel rebalance: a social channel that looked like it was losing money on last click can turn out to be profitable; see ROAS in SMA.
- Budget trade-offs change: cutting social because “it does not convert on last click” often brings brand Search conversions down a few weeks later.
The limits
| Limit | Consequence | Countermeasure |
|---|---|---|
| Only sees the journeys that are measured | Consent refusals, device switching, private browsing and offline conversions all escape it | Enhanced conversions, conversions API, offline conversion import, modelling |
| One model per platform | GA4, Google Ads, Meta and LinkedIn give different figures for the same sale; the sum across platforms exceeds 100% | One source of truth (GA4 or the CRM) to arbitrate between channels; the platforms to optimise within themselves |
| Correlation, not causation | A channel present in the journeys that convert is not necessarily the cause | Incrementality tests (geographic, control groups) for large budgets |
| Minimum volume | On a small account, GA4 falls back on rules | Group the conversions, lengthen the periods |
| Black box | The weights cannot be explained individually | Read the trends, not the decimals |
Reading and using attribution in practice
- Choose the source of truth: GA4 (paid and unpaid channels, data-driven model) or the CRM for long cycles. The ads managers are there to optimise within each channel, not to compare channels with one another.
- Compare the models: in GA4, Advertising, Attribution model comparison: last click against data-driven by channel. The gap shows the channels that are underestimated (often social) and overestimated (brand, remarketing).
- Look at the paths: Conversion paths in GA4: where social sits (start, middle, end) and how long the journeys are.
- Decide on trends: arbitrate budgets on three months of data-driven attribution, not on one week; test a cut in social budget and watch the effect on brand conversions at 4 to 6 weeks.
- Document: the model and the windows used in the dashboard, so that everyone reads the same thing; see the SMA KPIs and conversion tracking in SMA.
How GreenRed helps
Rather than juggling several tools, GreenRed's return on investment module brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
What is data-driven attribution?
A model that splits the value of each conversion between the channels in the journey according to their estimated contribution, calculated by comparing the journeys that convert with those that do not. It has been the default model in GA4 and Google Ads since 2023; it replaces last click, which gave everything to the final channel.
Why do Meta and GA4 not show the same number of conversions?
Each platform has its own model and windows: Meta claims every conversion within 7 days of a click or 1 day of a view, and models the losses; GA4 splits between all the measured channels using data-driven, on clicks only. A gap of 20 to 50% is normal. Use GA4 or the CRM to arbitrate between channels, and Meta to optimise within Meta.
Does data-driven attribution prove that a channel is profitable?
No: it measures a statistical contribution in the journeys observed, not a causal link. A channel can be present without being decisive. To prove incrementality you need a test with a control group or a geographic experiment; for an SME, comparing trends over three months and testing a budget cut are often enough.
Should you change the GA4 attribution model?
No, keep the data-driven model that is set by default; use the model comparison to understand the gaps with last click. The choice of model affects the attribution reports and the conversions exported to Google Ads, and therefore automated bidding; change it only if you know why.