Dr. Bratin Saha, former AWS executive who built SageMaker and Bedrock, discusses why most enterprise AI pilots fail to reach production. He shares frameworks for use case selection, the importance of last-mile specialization, and how NTT DATA's AI platform addresses governance and cost sprawl.
A short editorial from the VEONIB team on why this content matters.
Saha's core argument is that enterprise AI success hinges on last-mile specialization and governance, not just model capability. The pilot-to-production chasm is a systems problem requiring platforms, not point solutions.
Unlike typical AI hype, this interview provides a concrete framework for use case selection and emphasizes the underdiscussed role of change management and cost governance in AI deployment.
CIOs and enterprise architects should watch this to understand why AI pilots fail and adopt a platform-based approach to scaling AI with proper governance and value measurement.
The application of artificial intelligence technologies within large organizations to improve business processes and decision-making.
The critical transition phase where AI projects move from experimental testing to full-scale deployment and business value realization.
The process of customizing AI models with company-specific proprietary knowledge to make them effective in enterprise contexts.
Business processes where AI agents autonomously perform tasks, make decisions, and orchestrate other agents to complete work.
The policies, controls, and monitoring systems that ensure AI deployments comply with organizational rules and cost constraints.
The uncontrolled growth of AI-related expenses when multiple agents and models operate without centralized oversight.
The organizational process of transitioning workflows, roles, and human-computer interfaces to integrate AI effectively.
A monetization model where providers charge based on the business value delivered rather than time or resources consumed.
Why do most AI pilots fail to reach production?
Pilots fail because they lack last-mile specialization—the customization of AI with company-specific knowledge—and underestimate change management and workflow integration.
What is last-mile specialization in AI?
It is the process of ingesting proprietary company knowledge and context into AI models so they can operate effectively within a specific enterprise environment.
How should CIOs choose their first AI use case?
Select one or two use cases with moderate execution difficulty and high business value, then take them end-to-end to production before scaling.
Why can't frontier models like GPT-4 solve enterprise AI alone?
Frontier models lack access to proprietary company knowledge, and enterprises won't share trade secrets with model providers due to competitive and security concerns.
What is the role of an AI platform in the enterprise?
An AI platform provides standardized capabilities for deploying, governing, and monitoring AI agents according to company-specific policies and business objectives.
How will AI change IT services delivery models?
IT services will shift from staff augmentation and time-based billing to value-based pricing as AI agents automate workflows and deliver measurable business outcomes.
What is cost sprawl in AI and how can it be controlled?
Cost sprawl is uncontrolled AI spending from multiple agents and models; it can be controlled through platform-based governance, dashboards, and spending policies.
Why is change management critical for AI success?
AI requires reimagining workflows and human-computer interfaces, so employee buy-in and training are essential for adoption and value realization.
Should enterprises build AI on hyperscalers or on-premises?
Enterprises should build on top of existing hyperscaler infrastructure to leverage scalability and avoid reinventing foundational technology, unless operational constraints require on-premises solutions.
What skills should technologists focus on in the AI era?
Focus on foundational skills like problem-solving, collaboration, and leadership principles; the application of these skills will change but their importance remains.