AI voice agents on Vapi: inbound receptionist and outbound follow-up

The challenge

Warm leads from email outreach needed timely phone follow-up, and the firm's own line needed answering and routing, without hiring more front-office staff.

How it works

1. 01Inbound: answer and discloseGreets every caller, states it is an AI and that the call is recorded.
2. 02Inbound: identify and routeRecognizes returning callers, sorts prospects, clients, partners and vendors, and transfers in office hours.
3. 03Outbound: warm follow-upCalls warm email leads with context for each call: company, size, past touches and stated preferences.
4. 04Handles real conversationsGatekeepers, objections and "call me later" are handled; do-not-contact is honored at once.
5. 05Books or schedules a check-inChecks availability and books with an advisor, or sends a booking link and records a check-back month.
6. 06Structured call recordEvery call returns a structured outcome the dialer uses to schedule the next step.

What we delivered

• Never claims to be human

  AI disclosure, no hype or pressure language, and only approved facts and wording.

• Compliance guardrails

  Designed with compliance guardrails (AI disclosure, recording consent, calling hours, do-not-call).

• One core, no drift

  Both prompts are generated from a shared core plus a role file by a build script.

• Accurate details

  Emails spelled back, numbers read back, and resistance to prompt-injection attempts.

• Tool calls via webhooks

  Availability, booking, booking link, caller lookup and live transfer, all through n8n.

• Test plan included

  A 15-case test plan covering the main paths and edge cases before launch.

Stack

• Vapi
• GPT-4.1
• Deepgram Nova-3
• n8n webhooks
• Python
• Vapi API

Related service: AI agents & automation