Vapi Call Ops Report

Audit Vapi as a revenue and operations system, not merely as a call-volume dashboard.

Workflow

1. Identify the assistant/phone number/workflow, exact date range, timezone, environment, and available call fields.
2. Accept authorized connector/API data when available, or analyze CSV/JSON/transcripts/screenshots/logs. Never require a fixed Vapi schema; inspect the current fields first.
3. Build a call funnel from the evidence available:
   • attempted -> connected -> qualified -> desired action attempted -> action completed -> downstream outcome.
   • Desired actions may include appointment, transfer, message capture, quote/inquiry capture, or another business goal.
4. Segment by call direction, assistant/version, phone number/campaign, intent, disposition, ended reason, duration, time/day, and outcome where present.
5. Audit quality and failure modes:
   • hangups or premature endings
   • silence/latency/interruption problems
   • hallucinated or incorrect business information
   • failed tool calls, transfers, bookings, or CRM writes
   • loops, repetition, overtalking, poor intent recognition, weak objection handling
   • spam/non-customer traffic
   • transcription or language issues
6. Audit economics:
   • total and per-call cost when available
   • cost per connected call, qualified lead, completed action, and successful downstream outcome where denominators exist
   • expensive failure cohorts and avoidable token/model/telephony waste
7. Inspect transcripts selectively.
   • Prioritize failed high-value calls, anomalous long calls, repeated failure clusters, and representative successes.
   • Quote only short snippets needed to prove a finding.
8. Separate Vapi success from business success. A technically completed call is not a success if the customer failed to get what they needed.
9. Read references/report-contract.md and follow its output structure.

Rules

• Do not invent dispositions from transcripts when evidence is ambiguous; label inferred classifications.
• Do not treat average duration as inherently good or bad.
• Preserve privacy: avoid reproducing unnecessary personal information from callers.
• Prefer cohort evidence over isolated anecdotes.
• When Vapi platform behavior or field semantics matter and may have changed, verify current documentation before making platform-specific claims.

Useful triggers

Examples: "run a Vapi report", "audit yesterday's calls", "why is the voice agent missing leads?", "which call failures are costing us money?", "compare assistant versions".