Case Study
AVOD Monetization at Setplex - 3M+ RPS DSP and SSAI at Scale
Architected a distributed DSP and SSAI delivery stack handling 3M+ ad requests per second with sub-second latency - integrated with PubMatic, Aniview, and Kevel, with Broadpeak-based SSAI cutting ad-break latency by ~35%.
Outcomes
- Architected a distributed DSP system handling 3M+ ad requests per second with sub-second latency
- Reduced ad-break latency by approximately 35% via Broadpeak SmartLib SSAI optimisation, with improved fill-rate consistency
- Integrated with leading SSP/DSP partners (PubMatic, Aniview, Kevel) - expanding monetization channels across global markets
- Implemented Grafana/Prometheus observability for real-time latency, drift, and campaign KPI tracking
- Established CI/CD (GitLab + MLflow) supporting continuous retraining and environment versioning for production AI services
- Maintained >99.9% uptime across peak traffic windows on Azure cloud + microservices architecture
Context
Free ad-supported streaming (FAST + AVOD) is one of the fastest-growing segments in OTT, and the technical surface area is enormous. Server-side ad insertion (SSAI) for live FAST channels, programmatic demand across CTV/mobile/web, partner integrations across VAST/VMAP/ORTB, and region-aware monetization all have to work together - in production, on real devices, against real demand partners.
I led monetization workstreams across these layers as Setplex’s Head of AVOD, serving global FAST/AVOD audiences across web, mobile, and connected TV.
Scope
DSP at scale
Architected a distributed DSP (Python + Node.js + Redis microservices) handling 3M+ ad requests per second at peak with sub-second decisioning latency. Real-time bidding logic, frequency caps, and contextual signal evaluation all happen inside that budget.
SSAI optimisation
Directed backend optimisation of Server-Side Ad Insertion workflows using Broadpeak SmartLib, reducing ad-break latency by ~35% and improving fill-rate consistency. Tracked manifest issues (#EXT-X-PROGRAM-DATE-TIME, discontinuity edge cases), production/test tagging, and per-channel onboarding health.
Partner integrations
Drove integrations with leading SSPs/DSPs - PubMatic, Aniview, Kevel - clarifying VAST/VMAP/ORTB macros, identifiers, app store metadata, privacy fields, and CTV-specific validation constraints. Direct technical communication on cookie sync and identifier flows; ads.txt and app-ads.txt readiness across direct and programmatic demand.
Observability and CI/CD
Built CI/CD pipelines (GitLab + MLflow) supporting continuous retraining, automated rollback, and environment versioning. Deployed Grafana / Prometheus observability across the stack - latency, model drift, fill rate, campaign KPIs in real time.
Engineering coordination
Worked with engineering and product teams to optimise ad monetization workflows and viewer experience - ad break timing, live manifest requirements, compression considerations, and the boring rigor that determines whether monetization actually works.
What this work showed me
Monetization technology is a discipline of details that compound. A single mismatched macro, a missing identifier, a wrong region code, a manifest quirk on one player - any of these can quietly halve revenue without breaking anything visible. The work is about disciplined coordination across product, engineering, ad ops, and partners, with constant validation against the actual numbers.
Most of AdTech is unglamorous infrastructure work. The wins are real, but they come from boring rigor - not from clever tactics.