BANTify
Business problem
- After every discovery call, SDRs spend 10 or more minutes typing notes into Salesforce.
- Selling time disappears, records come out inconsistent, and AEs walk into meetings without the details that decide the deal.
- Conversation intelligence tools cost $80 to $150 a seat and still leave the CRM fields empty, so teams pay for recording and do the data entry anyway.
What I built
- An SDR pastes a transcript and clicks generate.
- The AI pulls out Budget, Authority, Need, and Timeline, scores the lead against a written rubric, and flags any competitor mentioned.
- A second AI audits every field against the transcript before the SDR confirms it.
- What lands in Salesforce is a complete Opportunity record with nothing invented and nothing typed.
Results
Architecture
A sales call transcript flows through three AI stages: extraction against a written rubric, a judge that validates every field against a transcript quote, and a comparison against past calls with the same company.
Everything then stops at a single human gate, the SDR's review checklist, which must be fully confirmed before the record can go anywhere.
Once approved, the record lands in two places at once: Firestore as a full audit-trail document, and Salesforce as a structured 13-field Opportunity ready for the Account Executive.
Key decisions & tradeoffs
Two walkthroughs
The same product, explained twice. One for the buyer who needs to see what it does, one for the engineer who needs to see how it works.
What it does, for the person deciding whether to buy it.
How it works, for the person who has to integrate it.