Problem
The operation’s video serving infrastructure — responsible for delivering VAST tags to campaign landing pages — was running on serverless architecture. It worked, but at growing scale, the economics and performance characteristics of serverless were becoming a liability.
Per-invocation costs were climbing with traffic volume. Cold starts introduced latency spikes that affected video load times on landing pages. And the operational complexity of debugging serverless functions under load was consuming engineering time that should have been spent on product work.
The challenge: migrate to dedicated hosting while campaigns are live, spending real money, serving real visitors. Any downtime means lost conversions. Any performance degradation means wasted ad spend.
System
The migration was engineered for zero downtime from the start, not hoped for.
Parallel deployment — the new dedicated hosting environment was built and tested alongside the existing serverless setup. Both systems served traffic simultaneously during the transition period, with the new system handling an increasing percentage of requests.
Monitored rollout — traffic was shifted gradually, with performance metrics (response time, error rate, first-frame delivery) monitored at each step. If any metric degraded, the rollout paused and the previous configuration absorbed the traffic.
Rollback plan — at every stage, the serverless system remained fully operational as a fallback. The migration was designed to be reversible at any point, not just at the beginning.
Performance validation — before shifting any production traffic, the new system was load-tested against historical traffic patterns. The team validated that dedicated hosting handled peak loads with better latency characteristics than the serverless setup.
Outcome
Zero downtime. Zero lost conversions. The migration completed without any campaign noticing the change.
Performance improved — dedicated hosting eliminated cold-start latency and provided more consistent first-frame delivery for videos on landing pages. Costs decreased — fixed monthly hosting replaced per-invocation serverless billing that had been climbing with traffic volume.
The migration produced a documented playbook for future infrastructure transitions: how to run parallel systems, how to monitor the cutover, and how to maintain rollback capability throughout. The same pattern has been applied to subsequent infrastructure changes.
Get case study breakdowns like this
One email when we publish. No spam, unsubscribe anytime. By subscribing you agree to our Privacy Policy.