Agentic Operation / Sales
Agentic RFQ-to-Quotation
Enquiries arrive as unstructured email and messages. Turning them into quotations means reading intent, matching products, checking stock and applying commercial policy — work that is slow, repetitive and easy to get subtly wrong.
The problem
Enquiries arrive as unstructured email and messages. Turning them into quotations means reading intent, matching products, checking stock and applying commercial policy — work that is slow, repetitive and easy to get subtly wrong.
The agent loop
- Trigger
- An RFQ arrives by email, WhatsApp or the web form.
- Context
- Customer record, item master, price lists, live stock, commercial history and past quotations.
- Agent
- Interpret the request, resolve the customer, match products and report a confidence score for each match.
- Policy
- Discount ceiling, credit hold, margin floor and match-confidence thresholds are evaluated deterministically, outside the model.
- Human checkpoint
- Sales authority approves only low-confidence matches or threshold breaches.
- Action
- Create the quotation, send it through the approved channel and schedule follow-up.
Implementation flow
Implementation flow / illustrative
- 01 / ExternalRFQ receivedEmail / WhatsApp / Web
- 02 / AgentParse and match productsFramework M
- 03 / SystemRead customer, stock and priceERPNext
- 04 / SystemCheck margin and creditPolicy engine
- 05 / HumanApprove exception if neededDesk
- 06 / SystemCreate quotationERPNext
- 07 / AgentFollow up and logFramework M
STATUS / READY — Discount ceiling, credit hold, margin floor and match-confidence thresholds are evaluated deterministically, outside the model.
- Tools the agent may call
- read_customersearch_itemsread_price_listread_stockcreate_quotation_draft
- Systems involved
- EmailERPNextFramework MWhatsApp Business API
- Framework M DocTypes
- CustomerContactItemItem PriceBinQuotationCommunication
- Industry
- Automotive
Reference architectures are illustrative; production design varies by system landscape, permissions, risk profile, data residency and model choice.
This page, for machines