AI-assisted RFQ processing
A human-controlled path from unstructured email to a reviewable commercial record draft.
Independently created. Contains no employer or client implementation detail, internal names or figures.
01Purpose
A pattern to adapt, not a blueprint to copy.
This reference architecture makes responsibility and handoffs visible before technology selection. It assumes a multi-system commercial environment where user experience, operational recovery and analytical consistency all matter.
02Logical flow
- 01Incoming email
- 02Classification
- 03Extraction
- 04Matching
- 05Human review
- 06CRM commit
03The agent design
Authority, tools, review and evaluation—designed per step.
The flow above becomes an agent only when every step has an actor, a tool contract, an authority level and an evaluation dimension. Drafts are reversible; commits pass through a person; the tools enforce it.
Designing an AI-assisted RFQ intake agent04Design rules
- 01Authority
Name the owner of every write, status transition and identifier.
- 02Boundaries
Exchange stable contracts; do not reproduce one system’s internal model everywhere.
- 03Recovery
Give failed work an owner, an observable state and a safe replay path.
- 04Human control
Place review where ambiguity becomes consequential—not at every step.
The boxes are the easy part. The architecture lives in the contracts, ownership and behavior between them.