Manager Queries

⚪ Planned (Phase 4) A manager should be able to ask the AI about any client, style, or order and get a consistent, complete, source-linked answer in seconds — instead of digging through months of email threads.

The query

The canonical question:

“What is happening with Jaspal style J26WGW153?”

Or variants:

“Show all RFQs waiting for quotation.” “Which styles have true cost but no quotation?” “Summarize latest client comments on this style.”

This is a query surface, not a chatbot. The queries are scoped templates, not free-form conversation. See Principles.

The challenge

A style lifecycle at VinMake takes 4–6 months from RFQ to bulk delivery. Along the way it accumulates:

Many email threads
Multiple client PICs
Multiple VinMake staff
Tech pack revisions
Sample comments
Costing updates
Negotiations
PO changes
Production updates
Delivery updates

No human can hold that in their head. The agent has to do strong retrieval and return a structured answer — not a wandering summary.

Required output structure

Every answer follows the same structure. The agent does not invent sections or skip sections. Consistency is the point.

Executive Summary
Current Status
Key Milestones
Pending Actions
Commercial Status
Product / Technical Status
Production / Delivery Status
Client Communication Timeline
Risks and Blockers
Recommended Next Steps
Source Links

The target is roughly 95% consistency in milestone structure across repeated queries. That target is aspirational and will be measured. To hit it, the agent needs:

Structured retrieval (not raw LLM recall)
A deterministic response template
Cached milestone extraction where possible
Source-linked artifacts on every claim

Example answer

Style: J26WGW153
Client: Jaspal
Internal PIC: Mai
Client PIC: Kaimook
Current Stage: True cost draft under review
Risk Level: Medium

1. Executive Summary
The style was received from Jaspal on [date] with tech pack and POM attachments.
Style record has been created. BOM draft is complete. True cost draft is
pending merchandiser review.

2. Current Status
- Style record: Created
- BOM: AI draft created
- True cost: Draft pending review
- Quotation: Not yet generated
- Client reply: Not applicable yet

3. Pending Actions
- Mai to review BOM quantities
- Sourcing team to assign actual button and shell fabric
- AI to generate quotation after true cost approval

4. Risks
- Material assignment incomplete
- No confirmed target price from client
- Button count needs human confirmation

5. Recommended Next Step
Review and approve true cost draft, then generate quotation with 15–20%
margin scenario.

Every claim links to its source: the original email, the tech pack, the BOM record, the true cost draft, the CutMake record.

What this is not

Not a chatbot. The agent answers from a scoped set of query templates.
Not unsourced. Every assertion links back to an artifact.
Not free-form. The structure is fixed. Sections don’t move.

Oversight Agent — push view of the same data
Principles — auditability and scoped chat
Roles / Manager — who this is for

​Manager Queries

​The query

​The challenge

​Required output structure

​Example answer

​What this is not

​Related