How to choose an attribution tool: a decision guide
Start from the decision and the data you hold. To cut or move a channel you need an experiment; a mix model needs years of weekly history; platform lift tests have entry thresholds. Otherwise start with GA4 and Shopify.
€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
Choose the method from the decision and the data you hold, then the tool. A budget move needs an experiment; a marketing mix model needs, per Robyn's documentation, "a minimum of two years of historical weekly data"; and Google's user-based Conversion Lift asks for "at least 1,000 observed conversions" and a "minimum campaign budget of $5,000 USD". If you hold none of those, start with GA4 and Shopify, and use a channel-level read to decide what to test.
What decision is the tool for?
A creative swap or a keyword change rarely needs an attribution tool: the platform's own report is enough for in-platform choices, even though platforms grade their own homework. A budget move needs evidence that a channel causes sales, and that is where methods split. In 15 Facebook experiments, observational methods "often fail to produce the same effects as the randomized experiments" (Gordon and colleagues, Marketing Science, 2019). Any tool that works from observed data needs an experiment behind it sooner or later, so the useful question about a tool is how it gets you to one.
What data does each method need?
Platform and GA4 credit. Nothing to buy. Google Ads makes data-driven attribution available to conversion actions "regardless of conversion or interaction volume", and recommends "at least 200 conversions and 2,000 ad interactions in supported networks within a 30-day period". It still works with less.
A platform lift test. Google says its user-based Conversion Lift "isn't available for all Google Ads accounts" and lists "at least 1,000 observed conversions" plus a "minimum campaign budget of $5,000 USD".
A holdout you run yourself. Orders by region from Shopify, and the ability to pause a channel in some regions. Experiments are noisy: in 25 retail and brokerage experiments, "the median confidence interval on return on investment is over 100 percentage points wide" (Lewis and Rao, Quarterly Journal of Economics, 2015).
A marketing mix model. Weekly spend and sales per channel, with spend that moved; Meridian's documentation warns that "insufficient variation in the media spend adversely impacts national models". Robyn recommends "1 independent variable: 10 observations". Meridian works an example of 12 channels, six controls and eight knots over two years of weekly data, which leaves "four data points per parameter", "too low to estimate the model reliably". It calls its own guidance "rough and directional".
Does your consent rate change the choice?
Yes. Robyn's documentation says multi-touch models are "very reliant on online signals, whereas MMM does not need user level data". If many visitors decline consent, path-based credit describes only the rest, and a holdout or a mix model, which work on aggregate counts, lose less. Meridian's documentation says MMM "uses observational data at an aggregate level that is privacy safe".
Which method fits your data?
Answer in order and stop at the first line that fits:
- You can pause a channel in some regions for at least your typical days to purchase. Run a holdout. It answers the cut-or-keep question directly, whichever tool you own.
- You cannot pause, but you hold two or more years of weekly spend and sales by channel, spend that moved, and few enough channels for the ratio to work. A marketing mix model is possible. Plan to calibrate it with an experiment; Robyn and Meridian both say to.
- Neither, and you run Google Ads at volume. Use its data-driven attribution for in-platform decisions and treat the credit as an estimate.
- None of these, but you have GA4 and Shopify orders. Reconcile the two first, then use a channel-level read from your GA4 export to decide what to test.
How do you check the fit before you pay?
Build the weekly table a mix model would need. One row per week, with revenue, spend per channel and any other driver you would control for, as far back as each platform goes. Then divide weeks by variables.
For illustration: 104 weeks, six channels, two controls and two time terms give 104 divided by 10, about 10 weeks per variable.
Pass: at least ten weeks per variable, which is the ratio Robyn recommends, and each channel's spend moved visibly at some point, through a launch, a pause or a seasonal ramp. Fail: gaps in the history, flat spend, or fewer weeks per variable than that. Merge channels or go to the next line of the list; do not buy a model your data cannot carry.
Later, if you hold GA4 but not the history or the ability to pause a channel, a causal attribution read like Causality Engine's works from a GA4 export and gives a next step per channel, with how to test it.
Sources, 30 September 2026: A Comparison of Approaches to Advertising Measurement (Gordon and colleagues, Marketing Science, 2019; abstract via IDEAS/RePEc); The Unfavorable Economics of Measuring the Returns to Advertising (Lewis and Rao, Quarterly Journal of Economics, 2015; abstract via IDEAS/RePEc); About data-driven attribution (Google Ads Help, 2026); Set up Conversion Lift based on users (Google Ads Help, 2026); An Analyst's Guide to MMM (Robyn documentation, Meta, read 2026); Amount of data needed (Meridian documentation, Google, 2026); About MMM as a causal inference methodology (Meridian documentation, Google, 2026).
Related answers
Frequently asked questions
How do I choose a marketing attribution tool?
Start from the decision, then the data. To cut or move a channel you need an experiment; a marketing mix model needs a long weekly history; platform credit is fine for in-platform choices. A tool is worth buying if it shortens the path to an experiment on the channel you would change.How much data does a marketing mix model need?
Robyn's documentation says a minimum of two years of historical weekly data and a ratio of 1 independent variable to 10 observations. Meridian's documentation calls its own guidance rough and directional and works an example where four data points per parameter is too low to estimate the model reliably.Is multi-touch attribution enough on its own?
Not for budget moves. In 15 Facebook experiments, observational methods often failed to match randomized results. Use multi-touch credit to see paths, and confirm any cut or increase with a holdout.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Keep reading
Terms in this article
- Attribution SoftwareAttribution Software measures campaign impact by tracking customer interactions across touchpoints. It assigns value to each channel, showing what drives conversions.
- Causal AttributionCausal Attribution uses causal inference to determine which marketing touchpoints genuinely cause conversions, not just correlate with them.
- Causal InferenceCausal Inference determines the independent, actual effect of a phenomenon within a system, identifying true cause-and-effect relationships.
- Confidence IntervalConfidence Interval is a statistical range of values that likely contains the true value of a metric. In marketing analytics, it quantifies uncertainty around estimates, indicating the precision of an outcome or causal effect.
- 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 MixThe marketing mix is the set of actions a company uses to promote its brand or product. It traditionally includes product, price, place, and promotion.
- 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.