Back to Website RETAIL AI PLANNING · 2026
M365 AI Knowledge Base Sync Architecture (Revised)
📣 Pre-Sales
Proposals / Marketing
AI HUB Feature Tools
Meeting Assistant
PPT Proposal Creation
Marketing Content Generation
Sales KB Q&A
Tender analysis · Creative generation tools
Citable sources · Multi-language translation
Pre-Sales Knowledge Base
• Product proposal library (life / health insurance)
• Marketing collateral & promo templates
• Competitor analysis & industry research
• Historical bid / winning case library
Retrieval (RAG Search)
• Proposal templates from similar clients
• Active product terms & selling points
• Client need profiles & competitor comparison
• Auto-translated foreign materials
Pre-Sales Dedicated Output
• Custom PPT proposal draft (auto layout)
• Meeting minutes + client to-do list
• Bilingual (CN-EN) product materials
• Marketing copy draft (sent to step ② dual-law review)
🔒 Filtered by role permissions · No unauthorized content
⚡ Efficiency Value
• Proposal creation: 3 days → 0.5 days
• Pre-meeting prep time ↓60%
• Juniors quickly reuse senior proposal expertise
• Significantly faster bid responses
📝 Underwriting
Underwriting / Policy Issuance
AI HUB Feature Tools
KB Smart Q&A
Document Translation
Meeting Assistant
Smart Data Query
Auto minutes & archiving for UW meetings
Source citations · Reinsurance doc translation
Underwriting Knowledge Base
• Policy terms library (filed versions)
• UW guidelines & rules manual
• Rate tables & actuarial basis
• Issuance policies & approval standards
Retrieval (RAG Search)
• Terms & interpretations for the product
• UW rules & past ruling cases
• Rate basis & discount policies
• Translated foreign reinsurance clauses
Underwriting Dedicated Output
• Terms interpretation (with source)
• UW advice & risk alerts
• Issuance meeting minutes & conclusions
• Quote basis notes (internal, no review)
🔒 Output after dual classification + permission check
⚡ Efficiency Value
• UW Q&A: hours → seconds
• Unified quote / issuance basis across branches
• UW conclusions traceable & reviewable
• New product training time cut in half
🛠️ After-Sales
Claims / Customer Service
AI HUB Feature Tools
Smart Q&A Assistant
Case Summary
Receipt Extraction & OCR
Voice Meeting Assistant
Auto summary after CS calls
Claims doc OCR · Smart pre-review
After-Sales Knowledge Base
• Claims guidelines & procedures
• CS scripts & FAQ library
• Renewal policies & conservation rules
• Historical cases & complaint library
Retrieval (RAG Search)
• Claims process & standards per product
• Resolutions of similar past cases
• Service scripts & complaint response plans
• Client data masked, viewed on demand
After-Sales Dedicated Output
• Claims assistance opinions (human review)
• CS response scripts & standards
• Case summaries & handling advice
• Renewal reminders & client notice drafts
🔒 Output only after masking + authorization check
⚡ Efficiency Value
• First-contact resolution ↑, transfers ↓
• New CS agent onboarding ↓50%
• Noticeably faster claims processing
• Unified service standards, fewer complaints
⚖️ Legal
Contracts / Compliance
AI HUB Feature Tools
Contract Doc Review
Contract Doc Comparison
Contract Interpretation
Legal Consultation
Review checklists · Annotated export
Sealed-version comparison · Diff highlighting
Legal Knowledge Base
• Contract templates & standard clauses
• Regulatory documents & law library
• Compliance policies & internal control manual
• Historical review opinions & case library
Retrieval (RAG Search)
• Applicable law texts & judicial interpretations
• Standard templates & baseline clauses
• Past reviews of similar contracts
• Regulatory penalty cases & risk alerts
Legal Dedicated Output
• Contract comparison diff report
• Review opinions & risk annotations
• Regulation search summary (with sources)
• Compliance review verdicts for marketing
🔒 Output by classification permission · Fully logged
⚡ Efficiency Value
• Contract review: 2 days → 2 hours
• Zero missed key risk clauses
• Faster regulatory response & reporting
• Legal expertise accumulated & reusable
🔒 Compliance Step ①
Data Compliance Check
Access / Retrieval Step
Protected End-to-End
• Entra ID authentication
• Business-line permission filtering
• Sees only own line's KB
• DLP sensitive detection
• Client info masking
• Cross-border data detection
• End-to-end TLS encryption
• Audit trail ≥ 180 days
PIPL / CBIRC
Classification · Minimum necessary
🔄 Channel Data Consolidation
Multi-platform sync layer · Read-only pull
📧 M365 Data Sync
SharePoint document libraries
OneDrive · Teams minutes
Outlook email archive
Graph API read-only · Permission inheritance
💬 WeCom
WeDrive docs · Chat archive
Calendar / meeting minutes
Knowledge mining from client group chats
Connector scheduled sync · Dedup & organize
🚀 Lark (Feishu)
Cloud docs · Knowledge base
Wiki / Bitable sheets
Meeting minutes · Collaborative organization
Open-platform API incremental sync
📌 DingTalk
DingDrive · Knowledge base
Online docs / spreadsheets
Approval flows · Announcement archive
Connector + scheduled task sync
🤖 AI Engine · RAG
Retrieval-augmented generation pipeline
① Query Understanding
Intent recognition
Query rewriting
② Embedding Vectorization
Semantic embedding
Vector index building
③ Vector Search
Top-K similarity recall
Permission filtering first
④ Semantic Rerank
Rerank fine ranking
Secondary classification filtering
⑤ Context Assembly
Prompt injection
Source metadata attached
⑥ LLM Generation
Cites original sources
Anti-hallucination · Traceable
⚖️ Compliance Step ②
Content Compliance Review
Output / Publication Step
Dual-law check in progress
• No absolute terms
• No exaggerated coverage claims
• No guaranteed-return promises
• Wording consistent with filings
• Output only after masking
• Outreach requires client consent
• Approval before external publishing
• Logs traceable & reviewable
Advertising Law × PIPL × CBIRC
Dual-law gateway · Dialectical review
Channel data consolidation & sync
Into business-line knowledge bases
🏗️ Technical Foundation · Cross-platform data consolidation (upstream sources of the KB)
📥 Unified consolidation pipeline (channel data → KB)
Read-only sync
ETL cleansing
Classification labels
Permission mapping
Vectorize & ingest
↓ Consolidated into pre-sales / underwriting / after-sales / legal business-line KBs (inheriting source platform permission schemes)
• Incremental updates: new / changed docs organized within 5 minutes
• No local plaintext · Full audit trail · Dedup & version control
• Classification labels: Public / Internal / Confidential / Top-Secret, 4-level control
☁️ M365 Technical Foundation
SharePoint
Graph API
Entra ID
Copilot
OneDrive
RAG Engine
DLP Protection
Audit Trail
Full TLS Encryption
Teams
• Data stored in Azure China (Beijing / Shanghai); no cross-border transfer
• AI Q&A not used for model training; enterprise knowledge stays inside
• Read-only API sync never alters originals · Inherits M365 permission filtering
⚖️ Dialectical Unity of Two Laws · Dynamic Balance
• Marketing efficiency and compliance baseline in dialectical unity; dynamic balance, two-way calibration
• Advertising Law governs wording × Data Law governs usage; two laws interlock and support each other
• AI output passes dual checks at the "compliance gateway": compliance first, publish later
• Review logs traceable; compliance drives development, development strengthens compliance
Advertising Law × Data Compliance Law × CBIRC industry regulatory requirements
