Detailed Use Case Estimates: AI Email Classification (Pillar 1)¶
This document provides a granular, bottom-up estimation for the AI Email Classification module (Phase 0/POC - 17 UCs).
Technical Pillars: - O365 Listener: Real-time polling and ingestion of mailbox signals. - AI Intent Engine: LLM-based classification of business intent (Sales vs Logistics). - Vector Ingest: RAG preparation for historical context retrieval.
| ID | Title | Pillar | FE (h) | BE (h) | QA (h) | Total (h) | Complexity Notes |
|---|---|---|---|---|---|---|---|
| EML-001 | Monitor Email Mailboxes | Ingest | 4 | 24 | 12 | 40 | OAuth2, token refresh. |
| EML-002 | Contextual AI Triage | AI | 10 | 60 | 30 | 100 | Multi-step reasoning & self-correction loops. |
| EML-003 | Automated Stream Routing | AI | 10 | 60 | 10 | 80 | Intelligent stream distribution & noise rejection. |
| EML-004 | Create Lead from Email | Bridge | 8 | 12 | 4 | 24 | Signature scraping. |
| EML-005 | Link Email to Lead | Bridge | 4 | 8 | 4 | 16 | Dedupe & thread logic. |
| EML-006 | Track Quote Milestone | Watch | 4 | 8 | 4 | 16 | SLA timers. |
| EML-007 | Lookup Order in ERP | Bridge | 2 | 10 | 4 | 16 | ERP API connector. |
| EML-008 | Set Dispatch Priority | Action | 4 | 6 | 2 | 12 | Manual override flag. |
| EML-009 | Send Dispatch Conf. | Auto | 2 | 10 | 4 | 16 | Webhook trigger. |
| EML-010 | Assign Task ID | Ingest | 2 | 6 | 2 | 10 | Unique sequence gen. |
| EML-011 | Update Task Status | Ingest | 4 | 6 | 2 | 12 | Sync to Kanban. |
| EML-012 | Send Escalation Alert | Watch | 2 | 8 | 4 | 14 | Manager notifications. |
| EML-013 | Multi-Entity Valid. | Logic | 4 | 20 | 6 | 30 | GST/CIN validation. |
| EML-014 | Pincode Route Plan | Field | 8 | 12 | 4 | 24 | Geofencing logic. |
| EML-015 | Campaign Tracking (AI) | AI | 4 | 12 | 4 | 20 | ROI attribution & intent map. |
| EML-016 | Outlook-SF Push | Plugin | 4 | 4 | 2 | 10 | Manual Right-click. |
| EML-017 | Entity Threading (AI) | AI | 10 | 30 | 10 | 50 | Knowledge graph link (Semantic). |
| Module Total | 96 | 316 | 98 | 510 | Adjusted for Agentic Orchestration depth |
Technical Allocation Summary¶
| Category | Effort (h) | Key Components |
|---|---|---|
| O365 Integration | 66 | OAuth2, Webhooks, Mailbox Polling |
| Contextual AI Core | 250 | Advanced reasoning loops, Self-Correction, Meta-Analysis |
| Data & Vector | 100 | Ingestion pipelines, Vector DB infra |
| QA & Reliability | 94 | Edge cases, accuracy hardening |
| TOTAL | 510 | Note: Training custom models (RoBERTa/DistilBERT) is covered by the 15% Project Contingency. |