Experimentation can drive better organizational decision-making, thereby significantly impacting company performance. In this post, Chang She describes TubiTV’s experimentation platform, including its experiment engine, UI (to create experiment configurations without managing JSON schema and to enable experiment tracking), automated analysis (including computation of metrics and significance for each metric), and QA methodology (which checks for failures like uneven splits, interaction effects, and biased exposures). He describes how this architecture increased experimentation velocity by enabling ML engineers to easily configure and deploy experiments; aligned experimentation culture with company strategic objectives; and afforded better, more agile decision making by enforcing consistency.