Attribution software: four types and what each can see
Four types exist: rule-based tools credit logged touchpoints, data-driven models compare converting and non-converting paths, mix models read aggregated spend, and experiments measure one randomised change. Pick the type that can see your question.
€99 once, excl. VAT. Full refund within 30 days, no questions asked. You keep the read.
By Joris van Huët, Founder & CEOPublished 5 min read
Four types of attribution software see four different things. Rule-based tools credit the touchpoints they log, data-driven models compare converting and non-converting paths, marketing mix models read aggregated spend and sales over time, and experiments measure only the change you randomise. Pick the type by the question before you compare vendors: GA4 has offered only three attribution models since November 2023, and Google's Meridian guidance recommends a minimum of two years of weekly data for a geo-level marketing mix model.
What can rule-based attribution see?
It sees the touchpoints a tag or pixel logs, and a fixed rule decides who gets the credit. GA4's first click, linear, time decay and position-based models are no longer available as of November 2023, which leaves data-driven, paid and organic last click, and Google paid channels last click. Shopify's marketing reports still offer last non-direct click, last click, first click, any click and linear. A rule can't see a visit that wasn't logged, and it doesn't compare a path with what would have happened without the touch.
What does data-driven attribution add?
It learns credit from data instead of a rule. Google says GA4's model evaluates converting and non-converting paths with a counterfactual approach and, for Google ad exposures, compares exposed users with similar users in a holdback group, training on data from randomised controlled trials. That is Google describing Google's own exposures. In Google Ads every conversion action is eligible, and Google recommends at least 200 conversions and 2,000 ad interactions in supported networks within a 30-day period so the model can find patterns. It sees what the platform and your property log, not another platform's impressions.
What can a marketing mix model see?
Aggregated spend and outcomes, with no user-level data. Meta's Robyn documentation describes no requirement for personal data or individual log-level data and no dependence on cookies or pixels. The price is history. Robyn's guide says a model needs a minimum of two years of historical weekly data, while Google's Meridian guidance gives two years of weekly data for geo-level models and three years for national ones, so the two differ on national models, and Meridian calls its guidance on the amount of data rough and directional. A mix model can't see individual journeys or a change inside a week.
What do experiments see?
The effect of one randomised change. In Google's Conversion Lift, people who see your ads form the treatment group and people who don't form the control group, and the difference in conversions is the lift. Google Research's 2011 paper on geo experiments does the same with randomly assigned regions. The result covers the channel and window tested, and Conversion Lift isn't available for all Google Ads accounts.
Which type answers which question, and how do you check?
| Type | Sees | Needs | Blind to |
|---|---|---|---|
| Rule-based | Logged touchpoints, credited by a fixed rule | A tag or pixel and tagged links | Unlogged visits, and what would have happened anyway |
| Data-driven | Paths of converting and non-converting users | Path data from your own property or account | Other platforms' impressions, and any path it isn't fed |
| Marketing mix model | Weekly spend and outcomes by channel | Years of weekly data, with spend that varies | Individual journeys, and changes inside a week |
| Experiment | The effect of one randomised change | A control group and a test window | Channels and periods you did not test |
Write your next budget decision as a question, find the row that can see it, and run the cheapest version. For "should I cut this channel?" that is a one-region holdout: lower that channel's spend a little in one region for a fixed window, then compare revenue there with a similar region. Pass: the gap is larger than the usual week-to-week swing between the two regions before the test. Fail: it isn't, so the test can't settle the question yet, and you extend it or pick a bigger region. Credit assignment alone is thin evidence for a cut: in 15 Facebook experiments, Gordon and colleagues found that observational methods often failed to produce the same effects as the randomised experiments (Marketing Science, 2019), though that study covers Facebook advertising only.
Sources, 30 September 2026: Get started with attribution (Google, Analytics Help); About data-driven attribution and About Conversion Lift (Google Ads Help); Marketing reports (Shopify Help Center); Robyn and the Robyn analyst's guide to MMM (Meta Marketing Science); Collect and organize your data and Amount of data needed (Google, Meridian documentation); Measuring Ad Effectiveness Using Geo Experiments (Vaver and Koehler, Google Research, 2011, research paper); A Comparison of Approaches to Advertising Measurement (Gordon, Zettelmeyer, Bhargava and Chapsky, Marketing Science, 2019, peer-reviewed).
Related answers
Frequently asked questions
What is the difference between rule-based and data-driven attribution?
A rule-based model assigns credit by a fixed rule such as last click. A data-driven model learns credit from your data; Google says GA4's model evaluates converting and non-converting paths. Both describe credit along logged touchpoints, not what a test would show.Which attribution models does GA4 still offer?
GA4 offers data-driven attribution, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based were removed in November 2023.Does marketing mix modeling need cookies?
No. Meta's Robyn documentation describes no requirement for personal data or individual log-level data and no dependence on cookies or pixels. It needs long history instead: a minimum of two years of weekly data, per Robyn's own guide.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Keep reading
Terms in this article
- Attribution ModelAn Attribution Model defines how credit for conversions is assigned to marketing touchpoints. It dictates how marketing channels receive credit for sales.
- Attribution SoftwareAttribution Software measures campaign impact by tracking customer interactions across touchpoints. It assigns value to each channel, showing what drives conversions.
- Control GroupControl Group is a segment of an audience intentionally not exposed to a marketing campaign, used to measure the campaign's true causal impact.
- CounterfactualCounterfactual is a hypothetical outcome that would have occurred if a subject had received a different treatment.
- IncrementalityIncrementality measures the true causal impact of a marketing campaign. It quantifies the additional conversions or revenue directly from that activity.
- Incrementality TestingIncrementality Testing measures the additional impact of a marketing campaign. It compares exposed and control groups to determine causal effect.
- Marketing Mix ModelingMarketing Mix Modeling (MMM) is a statistical analysis that estimates the impact of marketing and advertising campaigns on sales. It quantifies each channel's contribution to sales.
- Treatment GroupTreatment Group is the set of users exposed to a specific marketing intervention. Comparing this group to a control group shows the intervention's causal impact.