What billions of AI predictions taught Expedia before the age of AI agents

2026-07-06 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Expedia has developed a comprehensive set of ML and AI principles to guide the construction, deployment, and evolution of its AI systems, moving beyond basic predictions to agentic AI experiences. This framework addresses the challenge of building AI that scales and operates safely, not just works once. The principles are categorized into "Outcomes," focusing on aligning models to business impact, optimizing for return on cost, justifying complexity, and requiring both offline and online evaluation. "Design" emphasizes building systems that scale beyond individual teams by using shared foundations, treating data as a first-class product, prioritizing generality, minimizing manual rules, and ensuring reproducibility. Finally, "Trust" covers assigning clear ownership, adhering to governance, governing proportionally to risk, designing for fairness, privacy, and transparency, and implementing safe rollout, rollback, and continuous monitoring. These principles are translated into "Agentic Release" tollgates, integrating checks into the SDLC.

Key takeaway

For AI Architects or MLOps Engineers building scalable AI systems, you must operationalize governance beyond initial model deployment. Implement structured "Agentic Release" tollgates and integrate principles like clear ownership, risk-based governance, and continuous monitoring into your SDLC. This ensures your AI systems deliver lasting business value, scale safely, and maintain trust, preventing orphaned models and unmanaged risks as AI agents become more autonomous.

Key insights

Building scalable, responsible AI requires structured principles and operationalized governance across its lifecycle.

Principles

• Align AI efforts directly to business outcomes.
• Prioritize shared foundations for AI capabilities.
• Assign clear ownership for AI system lifecycle.

Method

Implement "Agentic Release" tollgates with recommended and required checks, integrating them into the software development lifecycle for designing, evaluating, approving, launching, and monitoring AI systems.

In practice

• Start with strong baselines before adding complexity.
• Require both offline and online model evaluation.
• Design deployments with rollback paths and controls.

Topics

• AI Governance
• MLOps
• Agentic AI
• AI Principles
• Model Evaluation
• Scalable AI Systems

Best for: MLOps Engineer, AI Architect, Director of AI/ML

Related on AIssential

• Production-ready agentic AI: evaluation, monitoring, and governance — Production agentic AI demands full-lifecycle evaluation, monitoring, and governance beyond mere functional accuracy.
• Enterprise AI Governance Beyond Model Risk: Why the Control Plane Is Becoming the Real Enterprise… — Modern AI governance must extend beyond models to the entire control plane, ensuring accountability and continuous oversight.
• Who Owns the Agent Once It Can Act? — Effective AI agent governance requires defining layered ownership and clear decision rights before autonomy scales.
• AI Governance Is the Strategy: Why Successful AI Initiatives Begins with Control, Not Code — Effective AI governance is an operational discipline enabling trust and value realization, not merely a compliance burden.
• How to build an agentic AI governance framework that scales — Agentic AI demands a new governance model that balances autonomy with oversight across the entire system lifecycle.
• CHAI’s framework divides governance into eight practical areas: AI policy, organizational structures, organizational resources, responsible lifecycle management, risk and impact assessments,... — AI governance must be an operational system embedded in workflows, not just abstract policies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.