How to Use GA4 Exports With Causality Engine: Setup Guide

Causality Engine runs on a GA4 export, not on a pixel or an integration. Export a window of GA4 traffic data as a CSV, add the Shopify orders export if you have one, upload both at app.causalityengine.ai, and the first causal read comes back in 5 to 10 minutes: per-channel incremental ROAS with confidence intervals, a platform-reported versus causal comparison for every channel, and a ranked budget reallocation you can export. The €99 read covers a 40-day window; Pro extends the look-back to your full history.

This guide covers what to export, what to check before you upload, and how to read what comes back. It is the same three-step method described on the how it works page.

Step 1: export from GA4

In GA4, open Reports, then Acquisition, then Traffic acquisition. Set the date range to the window you want read. For the €99 read, at least the last 40 days; for Pro, as much history as you have, because more history gives the model more natural variation to work with. Use the share icon at the top of the report to download the file as CSV.

Two checks before you close GA4. First, the date range should include weeks that differ from each other: weeks you scaled a channel, weeks you paused one, a promotion, a seasonal swing. The model reads causal effects from that variation, and a window in which nothing changed gives it little to read. Second, note how much of the traffic sits in Direct and Unassigned. That share is traffic GA4 could not resolve to a source, and it sets the coverage of anything estimated from the export. The Black Friday number nobody checks explains what to do with it.

Step 2: add the orders export, if you have one

If you run Shopify, export orders for the same window from the admin. Orders are the only figure in the stack that is a fact about money, and the read uses them as the ground the channel estimates stand on. GA4 alone is sufficient to run; the orders file makes the coverage arithmetic explicit. A native Shopify integration is on the roadmap and is not required.

Step 3: upload and wait

Sign up at app.causalityengine.ai, drag the files in, and wait. The standard run on a 40 to 90 day window finishes in 5 to 10 minutes. No OAuth grant, no tag in your theme, no DNS change, nothing touches your storefront.

What comes back

Three things, on one screen and in an exportable report:

• Incremental ROAS per channel, with a confidence interval. The interval widens where the data cannot tell, and that is information, not a defect. A channel whose interval includes zero has not been shown to add revenue at this data volume; it has not been shown to add nothing either.
• Platform-reported versus causal, per channel. The gap is the claim each platform makes beyond what the causal estimate supports. Summed across platforms and divided by orders it is the claim ratio, which The Price of Being Found treats as the single most useful number to put on the table when three systems disagree.
• A ranked reallocation. What to scale, what to cut, what to leave alone, in plain language.

What the model assumes, so you can check it

The model assumes the spend and sales data are complete and accurate, that the window contains enough natural variation to identify each channel's effect, and that no external factor unobserved in the data is moving sales in lockstep with one channel. It does not assume the platforms are honest, that the last-click rule is right, or that correlation is causation. Where an assumption is uncertain, the interval widens. The method is proprietary causal inference, counterfactual estimation on aggregated first-party data; it is not machine-learning attribution and not an LLM, though the Pro tier adds a chatbot for asking questions of your own results.

Before you pay

The read has a fit floor. Below roughly 5,000 euros a month in paid spend, or with fewer than 40 days of history, the intervals are usually too wide to move a decision, and the honest answer is to wait. Two of the five published customer cases on the case studies page are brands that were declined for exactly this reason. The €99 read includes a 14-day Pro trial and is refunded if the first read does not pay for itself in one reallocation decision.

What to do this week

• If you own the budget: export the last 40 days today and run the read. The result is a decision you can make on Friday.
• If you have to defend the number: run it on last quarter's window first, so the December review compares like with like.

The interactive demo runs the same model on a sample store with no signup, if you want to see the output before exporting anything.

Product facts as stated on causalityengine.ai on 9 September 2026: pricing, windows and refund terms are on the pricing page and change there first.