For years, Arik Sasson and Agnius Valaitis’s monolithic A/B testing data pipeline at Wix was a constant source of frustration, vulnerable to even the smallest coding error. When a major failure caused days of downtime, it became impossible to ignore the system's fragility. They recognized the need for a fundamental overhaul — not just incremental patches.

Guided by six core principles — ease of use, reliability, robustness, comprehensive coverage, flexibility, and transparency — they replaced the monolithic architecture with independent experiment-based pipelines, dynamically orchestrated by Airflow DAGs and executed using the Apache Spark engine. This new approach eliminated cascading failures, dramatically improved uptime, and significantly accelerated the experimentation cycle.

This modular redesign also unlocked new capabilities, such as AI-driven experiment evaluation. With standardized data interfaces, Arik and Agnius introduced “Evaluator” KPIs — flexible, interaction-based metrics that enable scalable assessment of AI products like chat and content creation tools.