Purchase events firing twice — reported ROAS was 2.4x, the truth was 1.2x

The situation

A US outdoor gear DTC brand had been running Meta ads for 14 months. Their dashboard showed consistent 2.4x ROAS and the media buyer was proud of the numbers. Revenue was $380K/month.

The problem

The Meta pixel was firing on the order confirmation page. A server-side CAPI implementation was added six months later by a different developer. Nobody set a deduplication key — both the pixel and the CAPI were sending purchase events independently. Every purchase was counted twice. True ROAS was 1.2x.

What we did

1. Audited browser network tab on order confirmation — pixel firing with fbclid. Checked CAPI logs — identical event firing 800ms later with no event_id
2. Implemented consistent event_id across pixel and CAPI using Shopify order ID as the deduplication key
3. Waited 7 days post-fix for Meta's dedup window to fully apply, then compared reported vs actual revenue in Shopify
4. Rebuilt bidding strategy from scratch using accurate ROAS baseline — paused 3 underperforming ad sets that had appeared profitable

Why it worked

When pixel and CAPI both fire without a shared event_id, Meta deduplicates nothing — it counts both. This is the single most common CAPI mistake. The brand had spent 6 months optimizing a 2.4x ROAS that didn't exist. The actual signal was 1.2x — barely above breakeven for their margins.