Overview
A global leader in pipe and steel manufacturing handles a high volume of customer technical enquiries, each requiring two detailed documents:
- Comment Sheet
- Raw Material Technical Specification (RMTS)
Producing these manually meant senior engineers spending days cross-referencing customer specifications against internal standards and filling complex templates, with inconsistent outcomes across engineers.
Customer needed a way to remove this bottleneck without losing expert judgment on compliance decisions.
TDG was built to automate the mechanical work clause extraction, gap analysis, and document rendering while keeping humans in control at every critical review gate.
Challenges
Business Challenges
- Throughput Complexities
Manual clause-by-clause document generation created a throughput complexity on high-value orders. - Compliance & Audit Risks
Inconsistent deviation handling created audit gaps and compliance risk on regulated specifications. - Lack of Traceability
There was no systematic trail linking disposition decisions to evidence or reviewers. - Workflow Delays
After a Comment Sheet was dispatched, there was no structured handoff to start RMTS work, causing delays.
Technical Challenges
- Multi-Format Document Parsing
Customer enquiries arrive as PDFs, Excel files, and DOCX documents, some scanned and some digital, requiring both cloud and on-premises parsing. - Retrieval Reliability
Semantic retrieval against large standards corpora risked stale or weakly grounded citations. - Complex Workflow Coordination
Coordinating a five-role, multi-gate workflow with rejection loops and escalations required a durable state machine, not a simple API.
Solution
1. AI-Powered Clause Extraction
AI-powered clause extraction parses and triages uploaded documents, with Claude producing structured clause objects that a Reviewer quality-gates before proceeding.
2. Semantic Gap Analysis
Semantic gap analysis runs each clause against a Qdrant standards knowledge base, generating grounded dispositions:
- Comply
- Deviate
- Exception
- Clarification
Weak citations are flagged automatically, while conflict clauses are routed to the Author for a precedence decision.
3. Comment Sheet & RMTS Generation Workflow
The platform automates the generation of both the Comment Sheet and the Raw Material Technical Specification (RMTS) while preserving the same governance and approval framework across both documents.
Once clause extraction and gap analysis are approved, the system auto-fills the Comment Sheet template, runs drift detection, and presents it for review. After the customer PO arrives, Claude generates the complete RMTS.
Both documents pass through the same sequential approval workflow:
- Author
- Reviewer
- Approver
before being sealed and dispatched, ensuring consistency, traceability, and compliance across the entire document lifecycle.
4. Enterprise Governance & Workflow Controls
The whole platform runs on strict role-based access across five personas, with every gate action logged for audit. Enquiries are automatically escalated to Admin after three rejections.
Impact Created
- Faster Turnaround
Document turnaround dropped from 2–3 days to hours. - Complete Auditability
Every gate action, edit, and decision is now fully auditable with actor, role, and timestamp. - Reduced Rework
Targeted AI regeneration on rejected clauses cut rework loops by roughly 60%. - Automatic Escalation
Automatic escalation after three rejections eliminates stuck enquiries. - Quantified Business Value
A team handling 50+ enquiries a month now spends its time on review and exceptions rather than generation.
Transformation Snapshot
| Feature | Before | After |
| Clause Extraction | Engineers manually extracted clauses line by line. | AI automatically extracts and structures clauses for review with 90% accuracy. |
| Document Generation | Templates were filled manually from scratch after each PO. | AI generates Comment Sheets and RMTS automatically. |
| Human Effort | Engineers spent most of their time generating documents. | Experts focus on review, validation, and exceptions. |
| Decision Tracking | Decisions were tracked over email threads. | Every action and approval is logged automatically. |
| Workflow Management | No structured handoff between Comment Sheet and RMTS. | End-to-end workflow with automated transitions and approvals. |
| Escalation Process | Enquiries could remain stuck indefinitely. | Automatic escalation after three rejections. |
Conclusion
TDG shows that AI's highest value in regulated industrial workflows isn't replacing expertise it's removing the mechanical burden of generation so experts can focus on judgment.
"From enquiry upload to sealed workflow every disposition grounded, every decision signed, every deviation justified. Not a faster document process. A better one."
