root@mindgraph:~# cat case-studies/microsoft-365-inbox-to-erp-quote-generator.md
Microsoft 365 Inbox to ERP-Validated Quote Generator
DOCUMENT: Microsoft 365 Inbox to ERP-Validated Quote Generator
INDUSTRY: Manufacturing
PROJECT: Sales Automation
CLASSIFICATION: CASE STUDY // PUBLIC RELEASE
SUMMARY:
An AI agent that reads incoming material requests, checks them against the ERP, and drafts the quotation before a person opens the email.
DETAIL:
Multi-format ingestion (OCR, Office attachments), live ERP validation per material, automatic reclassification, and multi-location quotation splitting — all triggered by email.
IMPACT:
CLIENT: Mumbai tools manufacturing conglomerate
02 / MANUFACTURING / SALES AUTOMATION
An AI agent that reads incoming material requests, checks them against the ERP, and drafts the quotation before a person opens the email.
Client: Mumbai-based tools manufacturing conglomerate
Tags: Manufacturing · Email Automation · ERP Validation · Document Generation · OCR
// Context
Customer material requests arrived by email — sometimes typed, sometimes a photographed handwritten list, sometimes a scanned PDF or spreadsheet — and every one required someone to read it, manually check whether each material existed in the ERP, work out which of the company's locations could fulfil it, and type up a quotation by hand. A single multi-item order could take a sales coordinator the better part of a day, and mixed orders — some items valid, some not — had no clean handling path.
// Our Solution
An AI agent that reads, checks, and drafts. It merges the email body with OCR'd images/PDFs and parsed Office attachments into one document, classifies the inquiry, and extracts every material number requested. Each material is validated live against the ERP; inquiries end up correctly labelled standard, mixed, or non-standard based on what's actually fulfillable — not just what was asked for. Valid materials are grouped by fulfilment location and turned into one quotation document per location, attached to an auto-drafted reply.
// Technical Details
- Multi-format ingestion: AI OCR for images and scanned PDFs (rendered at 300 DPI per page), document parsing for Word/PPT/XLS
- LLM classification (standard / mixed / non-standard) with automatic chunking and result-merging for long threads
- Live ERP validation per material line over a secured database connection
- Automatic reclassification when an inquiry contains both valid and invalid materials
- Location resolution mapping native material codes to fulfilment-location codes
- Multi-location quotation splitting — one document generated per fulfilment location
- Thread-aware processing: skips stale messages, cleans up duplicate drafts, auto-labels processed mail