AI-Powered Cognitive Decision Engine for Healthcare Operations

This case study covers the design and deployment of the Enigma Cognitive Engine, Document Management System (DMS), and Copilot layer. Operating within Vidal Health's TPA modernization, this platform provides medical and billing document understanding, rule evaluation, and human-in-the-loop (HITL) operator support for high-volume, regulated healthcare claims.

Healthcare operations depend on reading complex, multi-page medical records under extreme time pressure. Operators spent up to 10 minutes per case manually categorizing files, reconciling billing headers, and mapping room categories. Legacy processes suffered from high variance due to manual data entry, leading to inconsistent deductions and package mapping.

First Principles Analysis: The core problem was data extraction and validation mismatch. A simple OCR tool is insufficient because medical documents lack standard layouts. The product needed to build a cognitive proofing layer where every extracted fact could be traced back to its specific visual coordinates on the source document, providing trust indicators to operators.

Finance and quality teams faced high audit exposure due to tariff catalog deviations and room category caps being miscalculated. Operators experienced cognitive overload from switching between PDF viewers and core systems, leading to high override rates and low trust in automated recommendations.

We designed and layered a cognitive decision platform on top of existing systems:

• Document Chronology & DMS: A split-screen document viewer with chronological categorization and missing-document warning indicators.
• Reconciliation Proofing Grid: Surfaced OCR-extracted values side-by-side with original document fields, highlighting mismatch alerts.
• Allowed Amount Calculator: An interactive assistant mapping room categories and tariff package rules directly into the decision interface.

The AI system deployed Computer Vision and Natural Language Processing (NLP) models:

• Document Classifier: Classifies intake documents (Estimate bills, Pre-Auth forms, Clinical reports, ID cards) automatically.
• Structured Field Extractor: Extracts critical entities (VPA, bank accounts, admission dates, room categories, billing line items).
• Confidence Scoring & Indicator Engine: Flags low-confidence extractions, routing them directly to exception queues.

Samadhan led the product strategy and execution for the decision engine. He defined operator journeys, designed the confidence-band logic, established the human-in-the-loop exception routing policy, and authored specifications for the reconciliation proofing interface.

The Enigma platform successfully automated unstructured document extraction, reducing manual handling on high-volume documents. It established a reliable audit log for all manual overrides, captured before/after values with edit reasons, and minimized room category caps leakage.

Key Product Leadership Takeaways:

1. Operator trust is a product requirement: Presenting a recommendation without evidence leads to rejection. Surfacing confidence levels and direct document source links accelerates adoption.
2. Design overrides as first-class actions: Instead of blocking manual corrections, log overrides to capture rich data for model retraining.