At the recent VB Transform 2026 conference, NTT DATA AIVista CEO Bratin Saha shared solutions to overcome the 'last-mile' challenge of integrating advanced AI models into real-world business operations. He emphasized that building a comprehensive system around the model is more critical than the core technology itself. This is the deciding factor for the success or failure of AI projects in highly regulated and high-precision industries.
Context & Root Causes
Currently, the majority of enterprise AI projects fail during the deployment phase due to poor integration, a lack of deep domain expertise, and a lack of clear governance mechanisms. When used directly, leading large language models like GPT-5.5 or Opus 4.8 often fail to achieve the required accuracy for complex workflows, such as processing multinational insurance claims filled with handwriting and complex forms. Therefore, businesses need to customize AI systems for specific operational workflows rather than relying solely on the out-of-the-box capabilities of foundation models.
According to NTT DATA AIVista, fine-tuning models is no longer the top priority for enterprises during real-world deployment. Instead, they are focusing on leveraging 'tribal knowledge' and undocumented, non-digitized internal processes to shape AI behavior without exposing proprietary data to the outside.
Technical Analysis & Architecture
To address this bottleneck, NTT DATA AIVista's system focuses on three core technical components. First, it captures domain-specific business context in a way that AI can digest. Second, it orchestrates a multi-model ensemble—combining high-performance open-source models with large commercial models—to optimize operational costs.
Finally, the system establishes specialized guardrails to validate model outputs and automatically trigger reprocessing if errors are detected. This approach keeps the business logic within the surrounding system rather than inside the model itself, allowing enterprises to easily swap or upgrade foundation models in the future without disrupting the system.
Expert Insights & Perspectives
CEO Bratin Saha believes that technology is not the bottleneck in enterprise AI adoption. He noted: 'When you deploy AI in an enterprise, you are not just deploying a technology, you are moving an existing process from point A to point B.' The real value is generated by the transformed process itself, not by the underlying model that powers it.
Impact & Future Outlook
For Vietnamese enterprises looking to optimize AI investments, NTT DATA AIVista's approach offers a practical roadmap. Instead of over-investing in building or fine-tuning expensive LLMs from scratch, organizations should start by integrating AI into existing workflows to mitigate change management risks.
Once these processes stabilize, businesses can proceed to redesign entire workflows to maximize economic efficiency. This will be the dominant trend driving AI to deliver tangible business value rather than remaining confined to experimental pilot projects.