Conquering the Last Mile: How Enterprises Turn AI Investment Into Value

At VB Transform 2026, NTT DATA AIVista CEO Bratin Saha detailed how enterprise platforms bridge the last-mile operational gap for frontier models in regulated production. By wrapping foundation models in proprietary enterprise data, specialized guardrails, and domain-specific workflows, the platform targets the exact integration hurdles that cause most corporate AI deployments to stall.

The Last-Mile Dilemma in Regulated Production

Pouring capital into raw LLM parameter scaling no longer guarantees enterprise value. Most corporate AI projects fail during the implementation phase due to poor integration, domain specialization gaps, lack of governance, and unclear ownership of outcomes. Frontier architectures like GPT-5.5, Opus 4.8, and Fable 5 struggle out of the box with the dirty realities of corporate paperwork. Multinational insurance claims involve complex forms packed with handwritten text, overlapping checkboxes, and irregular layouts.

“These forms are pretty complex, often have handwriting, lots of checkboxes, and so on,” Saha noted during his conversation with VentureBeat CEO and editor-in-chief Matt Marshall. Out-of-the-box reasoning engines fall short on these tasks because foundational training data lacks institutional tribal knowledge. That undocumented expertise—routinely living exclusively in the heads of human workers—must be captured and engineered into the system.

Technology is rarely the primary bottleneck. Instead, success relies on a tripartite framework:

• Capturing proprietary enterprise context and making it consumable by AI.
• Deploying model ensembles so cost does not go through the roof.
• Enforcing domain-specific guardrails that check the model and force a redo when it gets something wrong.

System Architecture Over Model Monocultures

Fine-tuning foundation models ranks last among enterprise model-selection priorities, according to recent VentureBeat surveys. Organizations refuse to expose proprietary operational data or risk competitive advantages by over-relying on static fine-tuning cycles. AIVista bypasses this by building an intelligent harness around the model rather than altering the core weights.

“The last mile is about taking data that’s proprietary to you and using that to build a system around the model that can steer the model in the right way that can put the appropriate guardrails around it,” Saha explained.

This architectural choice preserves swappability. Engineering teams can route routine tasks to cheaper open-weight models while reserving expensive frontier reasoning engines for high-stakes decisions where errors carry heavy financial penalties. By pairing technical deployments with human subject-matter experts, the platform translates raw worker behavior into codified agent rules.

Balancing Seamless Workflow Embedding with Process Reimagination

Enterprises running mission-critical operations will not rip out a working process midstream. AIVista splits digital transformation into two successive stages to minimize change management friction.

“We are starting with embedding in the workflow because it’s easier change management,” Saha stated. Enterprises cannot afford catastrophic downtime during midstream process replacements. Once the AI agent establishes operational trust inside the existing pipeline, organizations advance to full workflow reimagination.

NTT DATA leverages its position as one of the world’s largest insurance third-party administrators to scale these neurosymbolic models and guardrail generators. While institutional tribal knowledge remains strictly bespoke for each client, the underlying orchestration platform scales horizontally across highly regulated sectors like insurance and advanced manufacturing.

In enterprise tech, value isn’t generated by the AI model itself. Value comes from successfully shifting a complex workflow from point A to point B.