The challenge
Every checkout change at Meridian started a debate. A/B tests ran in a separate tool, took a month and were read by one analyst.
The approach
Experiments moved into the same flags the engineers already used. Guardrail metrics stopped any test that hurt refunds or latency, and results went straight to the warehouse.
The result
Twelve experiments a month, a verdict in two weeks and an eleven percent lift in checkout conversion in the first year.