Atria With Causality Engine: Check What Radar Calls a Winner

The short answer: Radar can hand an ad an A for conversion in the same week the store's orders sit flat, and both are telling the truth about different things. Atria, the creative analytics tool for Meta and TikTok ad accounts, grades each ad from the last 30 days against a ROAS or CPA goal you set, using the conversions those platforms report (Radar). The grade says which creative wins on the platform's count. Whether the channel as a whole earned its sales is a separate question, and it is the one Causality Engine, a causal attribution tool that reads one GA4 export, answers. Its first finding is free at app.causalityengine.ai/start.

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

What to settle before the next "Scale these":

• Turn the goal into orders. For example, a CPA goal of €40 on €20,000 of monthly Meta spend expects 20,000 / 40 = 500 purchases. Hold that 500 against the month's store orders, not against Ads Manager.
• Ask the assistant you already pay for. With the Shopify connector, Raya can answer how last week's sales compare with what Meta reported, one of the example questions in Atria's own help article (Shopify connector).
• Find the third number. GA4's Paid Social row is a last-click count of the same ads' visits, once their links carry a paid medium.

Atria and the read, row by row

How Radar decides what won

The first time Radar opens, it asks for three things: a timeframe (30 days is suggested), a conversion metric (ROAS, CPA or a custom metric such as cost per trial) and the goal for that metric. It then sorts ads into Winners, High iteration potential and Iteration candidates, the last being ads with good early signs but too little spend to judge. It works on Meta and TikTok and needs a connected ad account with conversion data. Atria says the grading model is trained on over $12 billion of ad spend (Radar).

So the conversion grade is the platform's conversion, measured against your goal. Atria's help pages read on 11 November 2026 do not say which Meta attribution setting the synced figures use. Meta's own default is to show off-ad actions, such as a website purchase, by the ad set's attribution setting, crediting a purchase that follows a view or a click within a set number of days (Meta's attribution system). Assume Radar sees that unless Atria tells you otherwise.

Atria is open about what does not feed the grade:

• Shopify gives Raya orders, products, customers, discounts and returns, but "doesn't appear in Analytics dashboards, Radar or custom reports", and adds no blended MER or ROAS (Shopify connector).
• Triple Whale adds 73 metrics to reports, Pixel ROAS among them, with models such as Last Click and Linear, yet its data "is not used to rank creatives" (Triple Whale).
• Northbeam metrics work only in reports grouped by Ad name (Northbeam).

One loose end. The custom metrics article says a custom metric appears in Radar and may use Triple Whale or Northbeam metrics in its formula, on Business and Enterprise plans. Whether a Radar grade built on such a formula counts as ranking by Triple Whale's data, the two articles do not settle. Ask support before you rely on it.

What Atria-launched ads look like to GA4

Atria sends nothing to GA4. Its Launch feature publishes ads to Meta across several ad sets, and the launch guide covers Page, ad sets, creative, primary text, headline, description and call to action, with no field for URL parameters (launching ads). Tags therefore come from Meta. Ads Manager has a URL parameters field under Tracking, and Meta says traffic since 1 January 2024 may arrive with a campaign source of Facebook or Instagram and a medium of paid added for you (URL parameters).

That pair is enough. GA4 puts a visit under Paid Social when its source is on Google's list of social sites and its medium fits the paid pattern, which "paid" does (default channel group). TikTok is the one to watch: Google's channel descriptions name TikTok among the video sites behind Paid Video, so look in Traffic acquisition to see which row your TikTok visits fill. And keep every source value in one spelling per platform, as Google's URL builder guidance advises, so each platform reads as one source when you compare it with Atria's figures.

Worked example: two dashboards, one month

For example, with figures made up to show the sums and drawn from no store: a skincare shop books 2,416 orders in a month. Meta reports 1,388 purchases on €52,600 of spend. TikTok reports 702. Google Ads reports 655.

Platform claims: 1,388 + 702 + 655 = 2,745 orders, or 2,745 - 2,416 = 329 more than the shop shipped.

Radar is set to a CPA goal of €40. Meta's CPA is 52,600 / 1,388 = €37.90, under goal, so the best ads land in Winners. GA4's last click gives Paid Social 512 orders, a CPA of 52,600 / 512 = €102.73. Both are true by their own rule. Meta counts anyone who saw or clicked inside its window; GA4 counts the last tagged click and files the rest under Direct or other channels.

Radar is right about the ranking inside Meta. If one ad converts at twice the rate of another on Meta's count, it very likely is the better creative. What the grade cannot say is whether Meta as a channel caused 1,388 sales, 512 or something else. A causal read puts Paid Social's caused share next to the last-click rebuild of the same export, with a data-health score and a next step (how it works). When it disagrees with Meta, scale the winners in one region first and watch store orders there.

The free check, before any read

Two parts, neither paid. First the reconciliation: ask Raya the Shopify question above for the last full week, or do it by hand from Atria's Meta Ecommerce column preset and the Shopify orders report (adding metrics). Then a holdout, by hand: before scaling a Radar winner everywhere, raise its budget in one country and leave a similar country flat for three weeks. For example, if the scaled country's orders go from 610 to 702 and the flat one's from 540 to 551, the gap is (702 - 610) - (551 - 540) = 81 orders the extra budget can claim. That number stands without any tool.

When you want every channel weighed at once, Causality Engine, which I founded, runs on the GA4 Attribution paths export from the start page. The first finding is free and computed in the browser; the full read is €99 once and refundable within 30 days (pricing, checked on 11 November 2026).

What neither tool does for the other

Atria answers a creative question: which hook, offer or format to brief next. Causality Engine answers a budget question per channel, and it does not split by ad, creative or tag, so it will never say which of Radar's Winners to keep. It reads GA4 key events (one-off read), wants 12 months or more of export (how it works), calls a channel under 1% of credit on both methods too small to act on, and declines when spend is too small for a signal (homepage).

Radar's Winners tab carries the instruction "Scale these", grading is included while recommendations spend AI credits, and Atria's plans are sized by the ad spend they analyse, $500K a month on Core (pricing). A grade that sends more budget to the platform also moves an account toward the next spend band. That is how the plans are built, and it is reason enough to check the grade against the store first.

See also

• Atria: Raya, Radar, Inspo and the plans
• How to use Atria: labelling creative and turning winners into briefs
• Atria on the marketplace: Atria's affiliate program terms
• Causality Engine: what it reads, costs and declines
• How to use Causality Engine: one export through to a regional test
• Causality Engine on the marketplace: the affiliate terms, stated as the owner's
• Atria vs Foreplay: grading your own ads or saving others'
• What Direct traffic is hiding: where untagged buyers end up in GA4

FAQ

Can Atria push its Radar grades into Causality Engine?

No, and nothing travels the other way either. Causality Engine takes one file, the GA4 Attribution paths export; Atria's grades and Meta's synced conversions never enter it (integrations).

If I connect Triple Whale, does Radar stop using Meta's numbers?

Not for ranking. Atria says Triple Whale's data is not used to rank creatives; it shows beside your Meta and TikTok figures in reports grouped by Ad name (Triple Whale).

Which number should the Radar goal use?

My recommendation: set the CPA or ROAS goal from your margin, then treat Radar as a ranking inside the platform rather than a forecast of store profit. The break-even ROAS calculator gives the floor the goal should not drop below.