Recovering the 60% of conversions Meta couldn't see

The situation

The brand was running a mixed Meta + Amazon Ads stack with four active campaigns and strong top-of-funnel engagement — but flat revenue. Reported ROAS sat at 1.17× for three consecutive months despite rising spend.

What we found

Two compounding leaks:

1. Geo routing was broken. UAE campaign links were landing on amazon.in — pricing in INR, shipping rules that excluded GCC. Users hit the PDP, bounced, and Meta’s pixel recorded them as “engaged” (not converted).
2. Client-side pixel loss. iOS 17 + ad-block + aggressive tracking prevention was silently dropping ~60% of purchase events. Meta’s model was optimizing against the visible 40% — which skewed heavily toward lower-AOV repeat buyers.

What we changed

• Deployed server-side CAPI with deterministic event IDs (dedupe against the client pixel when it does fire).
• Rewrote the campaign link generator to respect ship_to_country from the product feed — no more INR pricing for GCC customers.
• Rebuilt creative rotation around AOV cohorts (so the model stops optimizing for trial-size repeat orders and starts optimizing for full-price new customers).
• Bridged Amazon Attribution tags into the same conversion graph, so the brand stopped paying Meta and Amazon for the same customer twice.

Outcome

Twenty-one days after the attribution bridge went live:

• ROAS: 1.17× → 3.4× (Meta-reported; confirmed in Shopify revenue).
• Attributed conversions: +148% — most of it was already there, just invisible.
• CAC: −41% at the same daily budget.

The lesson: before you scale spend, make sure you’re reading the board correctly. We didn’t make the brand’s ads better. We made the measurement honest.