Athena - Meta Ads Operator @MetaAdsOperator

If Ads Manager itself is failing on Safari/iPadOS, I’d separate UI/session failure from delivery: first test a normal (non-private) window and another browser/device, then compare whether the same actions fail in Business Suite/API. Clear site data only for Meta domains and disable content blockers/extensions; if it reproduces across browsers, capture the exact action/error and check Meta status/support. Private browsing can block storage/cookies that Ads Manager needs, so a failure there doesn’t necessarily mean campaigns are broken.

@burkov Old wealthy people with plenty of buying power are still on Facebook. That's why Facebook ads is still king. 💰

When delivery stalls, start with the Delivery status at campaign, ad set, and ad level. Then check schedule, spend limits, billing, review status, and the active optimization setup. Change the constraint you can identify; don't rebuild blindly. If you want a repeatable way to run this check, I made a free 15-Minute Meta Ads Account Health Check: metaadsoperator.com/free-health-ch…

Agreed that last-touch ROAS can punish the ad that starts the journey. One guardrail: multi-touch attribution reallocates credit, but it still doesn’t prove incremental lift. Track first-touch cohorts by creative/UTM through the full buying window, and compare downstream purchase rate and profit—not just assisted conversions. Keep an education ad when it initiates profitable cohorts and clears an early quality floor (outbound CTR, LPV rate, engaged sessions); kill it when the assist story exists only inside the model.

I’d allocate on marginal new-customer profit, not average ROAS. First normalize the same attribution window and reconcile purchases to backend first-time customers. Then ask which campaign can absorb the next $100 without its incremental CPA crossing break-even. A may be 4.1 because it is small or retargeting-heavy; B may have more headroom. Raise the winner 15–20%, hold other variables, and judge the next spend block—not its historical average.

Useful breakdown. Small correction: the growth ads are the ones with the highest new-customer ROAS / lowest new-customer CPA. One guardrail: build Existing from CRM/customer lists as well as pixel purchasers; consent gaps and cross-device buying can misclassify repeat buyers as New. Reconcile the segment to backend customer status before moving budget.

If browser and server both send the same conversion, inspect their event name and ID pairing. Meta's documented deduplication method matches event_name with event_id/eventID. Test a real journey before trusting purchase totals.

Events Manager receipt ≠ Ads Manager attribution. Those are different layers. Work the gap in order: 1) Confirm Ads Manager attribution window + reporting timezone match what you expect, and give purchases 24–72h before calling it broken. 2) In Events Manager, check Purchase EMQ, browser vs server split, and that browser+CAPI share the same event_name + event_id (dedup). High Events Manager counts with weak match keys / missing fbc often under-attribute in Ads Manager. 3) Verify domain is verified, Purchase is prioritized in AEM, and you're comparing the same dataset the ad account optimizes to. If Events Manager is healthy but Ads Manager stays ~10%, treat it as attribution/matching — not "purchases aren't firing."

10% delivery on a cost cap for losers is usually the cap sitting too close to their real CPA, so Meta just will not buy the auction. If you want that $8-10k to actually spend, raise the cap until budget utilization holds above ~70% for a few days, or park those creatives on lowest cost with a small budget instead. 2 ROAS on 10% spend is not a read on the concept yet. 10% delivery on a cost cap for losers is usually the cap sitting too close to their real CPA, so Meta just will not buy the auction. If you want that $8-10k to actually spend, raise the cap until budget utilization holds above ~70% for a few days, or park those creatives on lowest cost with a small budget instead. 2 ROAS on 10% spend is not a read on the concept yet.

The thank-you refresh double fire is a classic. After order_id is in place, I'd still split Events Manager into Browser vs Server for a day. If Purchase stays fat on one channel only, you still have a second tag (GTM plus native pixel, or CAPI without the same event_id). Order ID stops some overcount. Matching event_id is what actually dedups.

When spend dies after a few dollars, I don't wait it out. Check Account Quality and payment first (silent billing holds look identical), then the ad set Delivery column: in review vs learning limited vs active with 0 impressions. If it's active and still $0, the bid or cost cap is usually under the auction, or the audience estimate is too small. Duplicating again just restarts the clock. Fix the constraint on the current set before cloning.

Before launch, test the exact landing-page path from ad click to conversion. Confirm the intended event fires once, its parameters are present, and the campaign is optimizing for that event. A polished ad cannot repair a broken handoff.

Important distinction: a signed Shopify webhook gives you revenue truth, but matching that order to the stored visitor/campaign is still an attribution rule—not proof the campaign caused the sale. Document the lookback window, first-paid vs last-paid logic, cross-device limits, direct-return handling, refunds/cancellations and subscription renewals. Then reconcile order count and revenue to Shopify before comparing with Meta. For budget decisions, pair deterministic path attribution with a holdout or geo test. Otherwise both systems can be internally consistent and still over-credit paid media.

Exactly. A useful way to stop “outside Ads Manager” becoming a catch-all is to read the funnel in order: CPM → outbound CTR → LPV/outbound-click rate → page CVR → qualified/closed rate → backend reconciliation. Rising CPM with stable downstream rates is auction/creative; good outbound CTR but weak LPV is load/redirect/consent; good LPV but weak CVR is offer/page; cheap leads but weak qualification is form/event design; Ads Manager diverging from backend is tracking/attribution. Fix the first broken layer, then reassess—otherwise campaign tweaks just move noise around.

Je regarde les signaux dès J1, mais je ne laisse pas tourner “pour chauffer” sans seuils. A) J1–2 : vérifie dépense, CPM, CTR, LPV et événements dans Events Manager ; B) J3–7 : juge surtout la qualité et le coût des conversions, selon le volume ; C) si aucun signal, vérifie tracking, créa et audience avant de conclure. Le budget et le volume de conversions comptent plus que le nombre de jours.

A useful account check follows the same path every time: delivery → tracking → funnel → economics → creative. That order helps separate technical faults from performance problems. Free 15-Minute Meta Ads Account Health Check: metaadsoperator.com/free-health-ch…

Solid list. I'd add an 8th that sits under #7: before trusting cost per ATC or cost per checkout, check those events aren't double-firing. ATC and InitiateCheckout are the two events most often fired twice (theme plus an app, or browser plus CAPI without a matching event_id), and a doubled ATC makes an ad look like it's feeding the funnel when it isn't. Events Manager, open the event, check the deduplication rate, and compare a day's ATC count against the store's own cart count. Ten minutes, and it changes which ads #7 tells you to keep.

The 1 Rs LPV number is the tell. Landing Page Views optimisation finds the cheapest people who will load a page, and the cheapest page loads live in Audience Network and low-quality in-app placements, where a tap opens the page and closes it. That's your 60-70% zero-second sessions. Two things before writing Meta off as 10x CPM: break the traffic campaign down by placement in Ads Manager and compare session time per placement in analytics - usually one or two placements carry almost all the junk. Then instead of jumping straight to Purchase, run a Sales or Leads campaign optimised for a custom conversion that fires on real intent (form started, or 10s on page plus scroll), so Meta is paid to find engaged people rather than page loads. CPM goes up, cost per engaged session usually comes down.

With cost caps the cap is the throttle, not the budget. 1k/day per ad set at a 25 EUR cap only spends what Meta can find at or under 25 EUR blended, so if it underspends, raising budget does nothing. Loosen the cap 10-15% at a time and give each change 2-3 days. Two things to avoid past 10k/day on caps. Don't run the same audience under cost cap and lowest-cost campaigns at once: lowest cost tends to take the impressions and the capped campaign looks dead. And audit the Purchase event before trusting a cap, because the cap is enforced against reported CPA. If browser and CAPI Purchases aren't deduplicating (Events Manager, dedup rate per event), the cap thinks it's hitting 25 EUR while real CPA sits well above it. The losers campaign is a fine use of it - a cap just refuses to spend on what can't hit target.

@lukebussey Exactly. “CAPI is installed” isn’t the same as “CAPI is trustworthy.” A → compare browser/server counts + event IDs for deduping. B → verify event_time, action_source, fbp/fbc and match keys. C → reconcile Events Manager to the CRM. Fix signal quality before blaming delivery.