A call transcript is an unreliable source of commercial terms. This demo shows an architecture that treats it as one: probabilistic extraction on one side, deterministic pricing policy on the other, and a draft quote that never commits without a clean validation.
Synthetic transcripts and illustrative pricing. All scenarios are composite and invented for this demo. The extraction step here is a rule-based parser that stands in for an LLM or a conversation intelligence API, so the focus stays on the boundary between extraction and validation.
1. Call transcript
SeatsPlanDiscountTermBillingPayment termsRampTentative language
Pricing policy (editable)
2. Pipeline result
3. Draft quote (never committed)
4. Validation findings
Structured payload handed to the CPQ layer (JSON)
The contract between extraction and the CPQ layer.
Design principles
Extraction is advisory
Every extracted field carries its evidence span and a confidence level. Nothing is trusted because a model said it.
Policy is deterministic
Discount limits, term floors and payment rules live in code, versioned and testable, not in a prompt.
Conflicts stop the line
Two different values for one field are a finding, not a coin flip. The system routes to a human with both quotes of evidence.
The quote stays a draft
The output is a payload for a quote draft. Approval and commit remain with Deal Desk and the CPQ approval chain.
In a real environment
Data sources
Call transcripts can be requested from Gong by call IDs or by date range.
Field types and picklist values used for validation come from Salesforce describe, filtering out inactive values.
Discount and pricing policy comes from the organization's own price book and approval rules, so it is configured per client.
Limits and prerequisites
Gong API credentials are issued by a technical administrator.
Gong documents default limits of 3 calls per second and 10,000 calls per day for the company.
Salesforce API requests are pooled per org within a 24-hour period. For Enterprise Edition the allowance is 100,000 plus a per-license amount and any purchased add-ons, and exceeding it returns HTTP 403 REQUEST_LIMIT_EXCEEDED.
Linking a call to the right Opportunity must be confirmed in the target org. Gong lists a beta endpoint for manual CRM associations on calls.
How it would be validated
Label a sample of real calls with the quote terms a person would extract.
Measure missed and wrong terms before any draft quote is created automatically.
Keep the output as a draft that a person approves.
What this demo does not claim: the extraction here is a rule-based parser. A real version would use a language model and has to be evaluated on the organization's own calls.