Osmos’s ad infra helped a fashion retailer grow ad sales by 3.5x

Background: A Fashion Catalog Complex Enough to Need Better Ads

Osmos monetizes onsite ad inventory for one of India's largest fashion and lifestyle e-commerce marketplaces, spanning apparel, footwear, and accessories at national scale. Ad serving there had run for years on an in-house server that handled campaign management, reporting, and attribution as three disconnected systems.

Challenge: The Wall: Ad Infrastructure That Couldn't Keep Up

Shoppers navigate fashion catalogs almost entirely through filters: brand, size, category, price. Most ad infrastructure was never built to read that complexity. 

For this platform, it showed up as an infrastructure problem at the core: operations were fragmented, attribution didn't work, and reporting couldn't be trusted. Shoppers applying filters still saw ads with no relevance to what they'd filtered for, and fill rate stayed capped by the same aging system. Heading into its biggest sale of the year, the platform needed ad infrastructure that could hold up under peak load, not just perform on an average day.

Solution: Osmos’s Approach: Unified, Filter-Matched, Load-Tested

Osmos replaced the in-house ad server with a single unified platform running campaign management, reporting, and attribution together.

• Filter-matched relevance. Osmos mapped 40+ custom attribute filters to the platform's fashion taxonomy, so ad serving matched whatever filters a shopper had actually applied instead of showing generic inventory, directly lifting how often shoppers clicked through.
  
  
• Load-tested infrastructure. The new platform held ad serving latency at sub-50ms p95 throughout a roughly three-week flagship sale, even as fill rate climbed past what the previous server could deliver.
  
  
• Reporting and attribution ran on that same unified layer, so performance stayed measurable through the sale's peak traffic instead of breaking down under load.

Impact: Numbers That Speak for Themselves

Over a roughly three-week flagship sale in 2026, the new platform delivered:

• <50ms p95 ad serving latency, held throughout peak load during the sale
• 40% improvement in CTR compared to the non-sale period, as ad serving matched what shoppers had actually filtered for
• 2.5x fill rate increase over the previous in-house platform, with latency held at sub-50ms p95 throughout

‍

The marketplace now had a higher fill rate. The infrastructure held through the peak load and the platform delivered reporting and attribution with accuracy. All of this lined up to deliver 3.5x of the retailer’s  average daily sales during their biggest annual sale.