Financial services — AI document intake & routing
Insurance brokerage, 40 staff · Financial & insurance services
Client documents arrived by email, portal upload, and fax — then sat in a shared inbox until someone manually filed them in the DMS and updated the CRM. Misfiled policies caused compliance findings; brokers lost hours searching for the latest version.
The problem
Intake staff classified 200+ documents a week by hand — policy schedules, claims forms, ID checks, and correspondence. Naming conventions were inconsistent. CRM records lagged DMS by days. Compliance wanted an audit trail for every document touch; IT wanted something that worked with existing M365 and Dynamics.
Constraints
- — Documents contain PII — processing must stay in Australian Azure region
- — Classification accuracy must exceed 90% before production cutover
- — Brokers need override path — AI suggests, human confirms edge cases
- — Parallel run for two weeks — no big-bang switch
Approach
Step 1
Document taxonomy & samples
Mapped document types, DMS folder structure, and CRM field mappings with compliance and broker input. Built labelled sample set for model tuning.
Step 2
AI classification pipeline
Azure Document Intelligence extracts fields and classifies document type. Confidence scores route high-certainty docs straight through; low-confidence items queue for human review in a Teams-adaptive card.
Step 3
Routing automation
Power Automate orchestrates intake — email and portal triggers, classification, DMS filing with consistent naming, CRM record update, and audit log entry per document.
Step 4
Parallel run & cutover
Two-week shadow mode — AI classification compared to manual filing. Tuned thresholds, trained intake team on review queue, then cut over with rollback documented.
Deliverables
- Azure Document Intelligence model — tuned on brokerage document types
- Power Automate intake pipeline — email, portal, and manual upload
- DMS filing with consistent naming and CRM sync
- Human review queue for low-confidence classifications
- Audit log and compliance reporting export
Results
- Intake time
- ~70% of documents auto-filed without manual touch
- Misfile rate
- Compliance findings on document handling dropped to zero post-cutover
- Broker search
- Typical document lookup dropped from ~12 min to under 2 min
- Stack
- Document Intelligence · Azure OpenAI · Power Automate · SharePoint · Dynamics
“Documents land in the right place with the right name — and compliance can actually trace what happened.”
Facing something similar?
Similar outcome: AI Ops Production Sprint — $12,500 · 6 weeks.