Cutting cost per order by 46% for a D2C skincare brand

Monthly ad spend had grown from ₹4 lakh to ₹9 lakh over a year while order volume barely moved. The in-platform reports still showed a healthy return, which made the problem hard to see from inside the ad account.

Two things were wrong. The pixel had been firing a purchase event on the cart page as well as the confirmation page, roughly doubling reported conversions. And the entire account ran on three creatives that had been live for eight months, so frequency was climbing and cost per thousand impressions with it.

We rebuilt measurement first. The Conversions API was implemented server-side with proper event deduplication, the duplicate purchase event was removed, and platform numbers were reconciled against Shopify order data weekly until the two agreed within a few percent.

With trustworthy numbers, the real ROAS turned out to be 1.9×, not the 3.4× the dashboard had been reporting. That reset the conversation about what "working" meant.

Creative then became the focus. We moved to a weekly production cycle — three new angles per week across UGC-style video, problem-solution statics and founder-led explainer clips — with fixed kill criteria so losing creative was cut within seven days instead of running indefinitely.

The funnel was restructured into three clean layers with mutual exclusions, so prospecting stopped bidding against retargeting for the same users.

Prospecting ran on broad targeting with creative doing the segmentation. Retargeting was sequenced by recency and depth of engagement rather than a single 30-day window. A retention layer targeted past purchasers with replenishment timing based on actual product usage cycles.

Landing pages were rebuilt for the three highest-spend products, cutting load time from 4.1s to 1.6s and moving the key reviews above the fold.

Cost per order fell from ₹1,420 to ₹770 over five months while monthly spend held steady. Order volume more than doubled.

The compounding effect came from creative throughput: by month four the account had 40+ tested creatives instead of 3, and the winning set was being refreshed before fatigue set in rather than after.

Blended ROAS — measured against total revenue, not platform-attributed revenue — settled at 3.8×.