NTT DATA AIVista's Solution for the Last Mile of Enterprise AI Agents
Insights on how frontier models deliver real value in regulated production environments.
1The Last Mile Challenge
At VB Transform 2026, NTT DATA AIVista CEO Bratin Saha discussed operationalizing frontier models in regulated environments. He emphasized how reliability, context, security measures, and compliance determine whether AI delivers enterprise value. Currently, many corporate AI projects fail due to poor integration and lack of domain specialization.
2It's About the System, Not Just the Model
Saha stressed that enterprises should invest in the system around the model, not just the foundation model itself. The last mile work involves wrapping a frontier model with an organization's proprietary data, workflows, and guardrails. In regulated sectors like insurance, real-world challenges such as handwritten forms expose the limitations of off-the-shelf models.
3The Triple Formula for Success
According to Saha, success requires technology, domain expertise, and change management simultaneously. NTT's advantage lies in pairing AI experts with subject matter specialists. Interviews with client employees help uncover undocumented workflows. Technology alone isn't sufficient - acquiring domain knowledge remains the true bottleneck.
4Where the Financial Return Lies
For enterprises with large AI budgets, returns come from the work built around models rather than the models themselves. Saha emphasized that value is created by the workflow being transformed, not the assisting model. This paradigm demands a radical shift in investment priorities away from pure technology spending.
5Phased Transformation Approach
Rather than choosing between embedding AI into existing workflows or complete redesigns, NTT views these as sequential phases. They start with embedding for easier change management. For clients running mission-critical operations, this provides a safer starting point than disrupting working processes.
6Balancing Open-Source and Frontier Models
Keeping intelligence in the surrounding system rather than the model preserves swappability. Saha's team runs a mix of frontier and open-source models, using the latter for low-risk scenarios. Frontier models are reserved for cases where mistakes would be prohibitively expensive, particularly in regulated industries.
7Scalable vs Bespoke Solutions
In NTT DATA's platform, guardrail generation and neurosymbolic models scale across customers, while capturing each organization's tribal knowledge remains bespoke. The company's position as one of the world's largest insurance TPAs provides strategic advantage in acquiring this specialized expertise.
8Sustainable Competitive Advantage
Saha noted that 20 years of accumulated domain knowledge and trust can't be instantly replicated, representing NTT DATA's durable edge. In regulated industries, AI adoption depends more on such long-term expertise investments than on technology alone. Ultimately, successful enterprise AI implementations require disciplined systems thinking.