Causality Engine vs GA4 Data-Driven Attribution: One Export, Two Reads

The short answer: it usually starts with three tabs open. Meta's revenue, Google's revenue, the GA4 advertising report, and a store total smaller than any two of them added together. If your store runs GA4, spends on Meta, Google and TikTok, and has to decide next month's split between them, Causality Engine is the pick for that decision. It reads the Attribution paths export from the same GA4 property. One full read costs €99, and the first finding is free on the start page (prices from the pricing page, checked on 27 October 2026). Keep GA4's data-driven model running beside it. It costs nothing, and Google made it the default model in GA4 and Google Ads (Google Ads help).

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

Which of those tabs do you forward to finance? Data-driven attribution is Google's answer to who earned each sale. A causal read is a second answer, from outside the company that sells a share of the clicks.

Before you touch any setting:

1. Put a price on a wrong split. Take a store spending €18,000 a month on ads, for example, that sends 5% of it to the wrong channel. It misplaces 18,000 x 0.05 = €900 every month. GA4's model costs nothing to consult. One causal read costs €99, once, per the pricing page.
2. Open the comparison GA4 already gives you. Under Advertising, then Attribution models, the model comparison report sets data-driven next to paid and organic last click, with a % change column per channel (model comparison report). The channels that move most are where your budget question sits.
3. Check the window. The key event lookback window for purchases is 90 days by default. It can be cut to 60 or 30, and it applies to every model (attribution settings). A short window starves the channels that open journeys.
4. Decide now how you will know. Write down the channel you would raise or cut. Then change it in one region first, the test Causality Engine's how-it-works page attaches to every channel in a read.

The same paths, two models

Three places the reads part

Who built the yardstick

Google is open about its method, and that deserves credit. The attribution guide says the model compares users who saw an ad with similar users in a holdback group. It computes "the counterfactual gains of Google ad exposures by training on data from randomized controlled trials". For a Google Ads account, that fits the job. For a split between Google, Meta and TikTok, the counterfactual was trained where Google could run the experiment.

Causality Engine reads the paths in your own export and publishes its assumptions (how it works). They are complete GA4 data, enough converting paths to separate each channel, and no factor the data cannot see moving sales in lockstep with one channel. A channel under 1% of credit on both methods gets called too small to act on. It is not handed a verdict.

What happens to Direct

A big Direct row does not mean you broke something. The rules are doing what they say. Every GA4 model withholds credit from direct visits unless the whole path was direct (attribution guide). GA4 also files a key event under Direct when it has no path data at all (attribution paths report). So Direct becomes a row that collects buyers without explaining them. The causal read treats Direct as an arrival, not a cause. On the start page, one store's donated read covers 1 January to 13 October 2026 and €905,898 of purchase revenue. It gives 61.67% of Direct's last-click credit back to Paid Shopping, Organic Search and Cross-network. That is the company's published example, not a forecast for your store.

Settings that change the past

A new reporting model recalculates history in every key event report built on event-scoped dimensions. Session source and other session-scoped ones stay put (attribution settings). Conversions can also be reattributed for up to 7 days (attribution guide). Last quarter's screenshot and today's report can disagree. Nothing in the market changed. The ruler did.

What each costs, checked on 27 October 2026

GA4 is free of charge, per Google Analytics. Causality Engine lists one read at €99, refundable within 30 days. Pro, at €299 a month, bills monthly and is cancelled at the end of any billing month. Enterprise is a 90-day pilot on a custom quote. Prices exclude VAT, with no seat fees and no percentage of ad spend (pricing page).

The free model has an owner. Google also sells the search, Shopping and YouTube ads that sit in the same paths. The post that retired the rules-based models says Google made data-driven attribution the default in GA4 and Google Ads. It adds that, combined with automated bidding, the model improves performance (Google Ads help). That does not make the model wrong. It makes it one participant's view of a contest it also sells seats in.

Where an agent helps: on Pro, the chatbot over your own attribution data answers follow-ups, such as which channel lost most share once Direct was split (pricing page). The human check stays with you. Open the read's receipts, which tie every number to the export, and find those rows yourself before the answer reaches a budget email.

Choose Causality Engine if your job is

• Splitting one budget across Meta, Google, TikTok, email and organic, with each channel's caused share set against its last-click share.
• Explaining a Direct row large enough to distort every other channel's credit.
• A quarterly second opinion on GA4 with no access grant, no tag and no change to your property.

Choose GA4 data-driven attribution if your job is

• Reading credit for Google Ads, YouTube engaged views and app campaigns inside Google's own tools, at no cost.
• Daily reporting where fractional credit by source, medium and campaign is enough to act on.

Background on the product: the Causality Engine glossary entry. For pulling the file, see how to export GA4 data for attribution; for models in a larger organisation, marketing attribution in the enterprise.

Run the free comparison, then the read

Start with what GA4 already shows. Note the three channels that move most between data-driven and last click. Then drop the Attribution paths CSV on the start page; the preview reads the file in your browser and does not upload it, per the privacy policy. If the preview and GA4 agree on those three channels, the answer cost you nothing.

If they disagree, run the full read in Causality Engine, which I founded, from the start page: €99, back in 1 to 2 minutes, with a health score and a next step per channel, as the how-it-works page describes. Pro fits 8 or more channels with budgets that move weekly, per the pricing page. If spend is too thin to separate channels, the read says so. The case-studies page includes a brand told its gap sat inside the confidence interval and advised to come back when paid spend scales.

FAQ

Does Causality Engine need access to my GA4 property?

No. Its GA4 page says it works from the standard export, with no account permissions and no tag. You download the Attribution paths report yourself with Share this report, as the attribution paths article describes.

Which attribution models does GA4 still offer?

Three: data-driven, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based went in November 2023, per Google's attribution guide.

Why do the two disagree?

Both describe a counterfactual. Google's is built per key event, from your paths and from trials of Google ad exposures. Causality Engine's comes from the converting paths in your export, with Direct split back and a health score per channel (how it works). So they part most where Direct is large, or where non-Google channels open the journey.