Causal Attribution for Shopify Brands Using GA4

A Shopify brand already holds the two files a causal read needs: the orders export, which is the only figure in the stack that is a fact about money, and the GA4 export, which is what the tracking could see. Shopify's marketing reports and the ad platforms each apply their own attribution model and window to the same orders, and none of them answers what a channel caused. Read the two exports together and you get per-channel incremental ROAS with confidence intervals, with the coverage of your tracking stated instead of hidden.

Three scoreboards, one order

A customer clicks a Meta ad on Tuesday, comes back through a branded Google search on Thursday, opens a Klaviyo email on Saturday and buys. Meta counts a purchase inside its 7-day click window. Google counts a conversion. Klaviyo credits the email. Shopify's marketing report applies its own model and window, set in the admin, and credits one of them. GA4 applies data-driven attribution to the sessions it could resolve. Five reports, one order.

The Price of Being Found calls this a category error rather than a reconciliation problem. The platforms report conversions they can associate with their own inventory inside their own windows. Analytics reports sessions it could resolve to a source. The store reports orders. Only the last is a fact about money, which is why the orders export is the ground everything else has to stand on.

The two numbers to compute before anything else

Coverage. GA4 purchase conversions attributed to any named source, divided by Shopify orders for the same dates. The gap is traffic GA4 could not resolve: consent declined, in-app browsers, cross-device journeys, stripped referrers. The book's own census at one company found 48.6% of sessions with no resolvable source, which is one company's data and not a population estimate, but it is the first query to run because every cost figure computed from GA4 is the real figure divided by this number.

Claim ratio. Meta's, Google's, TikTok's and Klaviyo's claimed conversions summed without deduplication, divided by Shopify orders. Above 1 for almost every multi-channel brand. The one-hour claim ratio audit is the procedure, and the Shopify orders export is the denominator in both.

What a causal read adds

A causal read takes the GA4 export and the orders export and estimates each channel's incremental contribution from the natural variation in the window: weeks a channel was scaled, weeks it was paused, promotions, seasonality. It returns incremental ROAS per channel with a confidence interval, a platform-reported versus causal comparison for every channel, and a ranked reallocation. It does not need a pixel, an app in the theme or an OAuth grant, so it runs alongside whatever Shopify apps are already installed and conflicts with none of them. How to use GA4 exports with Causality Engine has the export steps.

Two limits, stated up front. It is an observational estimate, not an experiment, and it says so in its interval. And it has a fit floor: below roughly 5,000 euros a month in paid spend or 40 days of history the intervals are usually too wide to move a decision. Two of the five published customer cases were declined on that floor.

What the Shopify apps cannot do

Pixel-based dashboards (the cheapest attribution tools for Shopify brands are compared elsewhere) install a script on every page and allocate credit across the touchpoints they observe. That is multi-touch attribution: a rule for sharing credit among journeys the pixel saw. It is not a counterfactual, and it is bounded by the same coverage problem GA4 has, because a pixel cannot see a consent-declined visitor either. The distinction matters most for the channel closest to the purchase, retargeting and branded search, which collects the most credit and, in the experiments the book quotes, tends to cause the least.

When to run a holdout instead

For the biggest channel, an observational read is the map and a holdout is the territory. Qualify it first: spend share times honest return has to clear the smallest lift a geo test can detect at your scale, about 8.3% for a typical DTC brand on the standard design. If it clears, hold the channel out of a slice of regions for eight weeks with the pre-period fixed, and let the causal read cover the other channels between anchors. Incrementality testing for ecommerce is the playbook.

What to do this week

• If you own the budget: export Shopify orders and the GA4 traffic acquisition report for the same 40 days, compute coverage and the claim ratio, and then run the read. Three files, one afternoon.
• If you have to defend the number: put the orders count on the first slide and every platform figure after it, with the claim ratio between them.

The interactive demo runs the read on a sample store with no signup.

As of 9 September 2026. Coverage, the claim ratio and the three-systems identity are from The Price of Being Found (Edition 2.10), Chapters 8 and 9, with the book's caveats. Product facts as stated on causalityengine.ai on the same date.