D2C · Skincare & Beauty

Blended ROAS of 1.8× against a 2.5× target. ₹25L/month in Meta spend across 4 agencies managing different campaign types simultaneously. Three separate pixel events all firing for the same purchase event, double-counting inflating the reported ROAS. No hold-out testing meant the team couldn't tell if Meta was driving conversions or just claiming credit for organic intent. Post-iOS 14, signal quality had fallen to 2.8/10, causing the algorithm to misfire on audience optimisation.

Fixed the pixel event duplication first, standardised to a single Purchase event with deduplication keys across browser and server. Built server-side CAPI with GTM, passing Shopify order data server-to-server, lifting the match rate from 31% to 74%. Ran a geographic hold-out test over 3 weeks to measure true incrementality. Rebuilt the campaign hierarchy into three layers: prospecting (CAPI Advantage+), retargeting (sequential creative), and retention suppression. Synced Klaviyo purchase data back into Meta as a custom audience seed for LTV-weighted lookalikes.

True incremental ROAS via hold-out: 4.3× (was reporting 1.8× but was likely lower than that). Signal quality score improved from 2.8 → 8.4/10. Cost per purchase fell 39%. ₹18L/month of spend was reallocated from low-ROAS broad audiences to LTV-positive cohorts identified from Klaviyo purchase history. Retention revenue (Klaviyo-attributed) was isolated from paid acquisition metrics for the first time, making the P&L readable.