Using AI Voice for Inbound Calls: Receptionist, Qualifier, Appointment Setter

Why inbound AI is underused

Most discussion of AI calling focuses on outbound. But inbound AI deployment often delivers higher ROI than outbound:

• Inbound calls already have intent (caller chose to call you)
• Higher conversation engagement
• Lower opt-out / hostility risk
• Less compliance complexity (no TCPA outbound rules)

Yet most businesses still use traditional voicemail or hire receptionists for inbound. AI sits idle.

Here's how to deploy inbound AI that handles the load.

---

Use cases for inbound AI

1. After-hours answering

Calls outside business hours. AI greets, takes a message, schedules a callback, or handles common questions.

2. Receptionist replacement

For small offices that can't afford full-time receptionists. AI handles initial routing, screens calls, takes messages.

3. Lead qualifier before routing

Caller dials your sales line. AI qualifies (interest, budget, decision-making) before routing to human rep. Human only takes pre-qualified calls.

4. Appointment scheduler

Caller wants to book an appointment. AI accesses calendar, finds slot, books, sends confirmation.

5. FAQ handler

Common questions (hours, location, services, pricing range) answered by AI without human involvement.

6. Order status / account lookup

Authenticated callers can check order/account status via AI without waiting for human.

---

Architecture: VAPI inbound assistant

Setup

In VAPI:

1. Create an inbound assistant (separate from your outbound)
2. Configure system prompt for inbound context
3. Assign a phone number (the public-facing one)
4. Configure routing logic and function calls

Prompt structure

Different from outbound. Inbound prompt:

You are Sarah, the AI assistant for Smith Solar. You answer the main business line.

When someone calls:
1. Greet warmly: "Thanks for calling Smith Solar, this is Sarah. How can I help?"
2. Listen to their request
3. Based on what they say, take appropriate action:
   - Sales inquiry → qualify briefly, transfer to sales or book appointment
   - Existing customer support → ask for account/ticket info, transfer to support
   - General question → try to answer; if can't, transfer or take message
   - Wrong number → politely inform and end
   - Hostile / urgent → de-escalate, transfer to human

Functions available:
- transfer_to_human (with team selection: sales, support, billing)
- book_appointment
- check_business_hours
- take_message

Tone: warm, professional, helpful. Don't be corporate or stiff.

---

Inbound flow patterns

Pattern 1: Greet → triage → route

Caller: "Hi, I'm calling about solar panels."
AI: "Of course! Are you looking to learn about solar for the first time, or following up on something?"
Caller: "First time, I want to know about pricing."
AI: "Got it. Pricing depends on your home — let me get you set up with a consultation. We have openings tomorrow at 2pm or Wednesday at 11. What works better?"
[books appointment]

Pattern 2: Greet → qualify → escalate

Caller: "Hi, I'm a customer and I have an issue."
AI: "Sorry to hear that. Can I get your account number or the email on your account?"
Caller: "[provides info]"
AI: "Thanks. Let me connect you with our support team."
[transfer to human]

Pattern 3: Greet → answer → end

Caller: "Hi, what are your hours?"
AI: "We're open Monday through Friday, 9am to 6pm Pacific. Saturday 10-2. Closed Sundays. Anything else I can help with?"
Caller: "No, that's it, thanks."
AI: "Have a great day!"
[end call]

---

Function: book_appointment

Inbound's superpower: caller wants to book, AI books in real-time without involving humans.

{
  "name": "book_appointment",
  "description": "Book an appointment in the calendar",
  "parameters": {
    "type": "object",
    "properties": {
      "datetime": {"type": "string", "description": "ISO 8601"},
      "contact_name": {"type": "string"},
      "contact_email": {"type": "string"},
      "contact_phone": {"type": "string"},
      "appointment_type": {"type": "string"},
      "notes": {"type": "string"}
    }
  }
}

When AI calls this function:

1. Webhook to your Make.com / n8n
2. Check calendar availability for the requested time
3. If available, create the appointment in GHL/Cal.com/Calendly
4. Send confirmation email/SMS to caller
5. Return success to AI: "Booked!"
6. AI confirms with caller: "You're booked for Tuesday 2pm. You'll get a confirmation email shortly."

---

Function: check_business_hours

Simple but important.

{
  "name": "check_business_hours",
  "description": "Get business hours for a specific day or current status (open/closed)",
  "parameters": {
    "day": {"type": "string", "description": "Day of week or 'today'"}
  }
}

AI uses this to answer "are you open?" without hardcoding hours into prompts (which become stale).

---

Function: take_message

When AI can't help and humans are unavailable.

{
  "name": "take_message",
  "description": "Take a message and notify the team",
  "parameters": {
    "type": "object",
    "properties": {
      "name": {"type": "string"},
      "phone": {"type": "string"},
      "subject": {"type": "string"},
      "details": {"type": "string"},
      "urgency": {"type": "string", "enum": ["urgent", "normal", "low"]}
    }
  }
}

AI:

1. Gets details from caller
2. Calls function
3. Webhook sends Slack message to team
4. Confirms to caller: "I've passed your message along. Someone will reach out within [timeframe]."

---

Setting up time-of-day routing

Different behaviors for different times:

Business hours

• AI greets, qualifies, transfers to humans for complex calls
• Books appointments live
• Routes urgent calls to on-call

After hours

• AI greets, indicates after-hours
• Books appointments for next business day
• Takes messages for non-urgent
• Transfers urgent (e.g., emergency services) to on-call number

Implementation

In your VAPI assistant or Make.com handler, check current time vs. business hours. Adjust prompt or function behavior accordingly.

---

Voicemail vs. AI

If you currently use voicemail:

Voicemail problems

• Most callers don't leave messages (10-15% leave-rate is typical)
• Messages often vague or incomplete
• Slow callback (you call them later, they're not available, phone tag begins)

AI replacement benefits

• 100% of callers engage (no need to leave message)
• Structured info captured
• Real-time appointment booking
• Better caller experience

For most businesses, replacing voicemail with AI is a clear win.

---

When NOT to use inbound AI

1. Highly emotional/sensitive calls

Mental health hotlines, hospice services, crisis lines — humans should answer immediately. AI feels callous.

2. Industries where customers expect human-first

Premium concierge services, luxury hospitality. AI breaks brand promise.

3. Highly regulated industries with strict scripts

Some financial services, certain medical contexts have regulatory requirements that complicate AI deployment.

4. When call volume is too low

If you get 5 calls/week, you don't need AI. Just use a phone.

---

Cost vs. receptionist

Receptionist (full-time)

• Salary: $35k-$50k/year
• Benefits: +30%
• Total: $45k-$65k/year for one person

AI receptionist

• Setup: $5k-$15k one-time (depending on complexity)
• Ongoing: $200-$1,000/month for typical SMB volume
• Annual: $2.4k-$12k after first year

For most SMBs, AI is 5-10x cheaper than a receptionist. With proper deployment, the AI handles 70-90% of calls without escalation.

But: if your business needs human touch on every interaction (concierge style), the AI savings come at brand cost. Calculate both.

---

Common pitfalls

1. Generic greeting

"Hello, how can I help you?" sounds corporate. Personality matters.

2. Slow function calls

If "book_appointment" takes 8 seconds to confirm, caller wonders if AI hung up. Optimize function call latency.

3. No fallback for unknown questions

AI gets a question it can't handle. Without graceful fallback, conversation feels broken.

Solution: explicit "if I can't help with this, I'll connect you with someone" path.

4. Forgetting peak hours

Inbound traffic spikes (Monday morning, after marketing campaigns). Test under load.

5. No analytics

Are people getting their needs met by the AI? Track:

• Call resolution rate (% of calls completed without human transfer)
• Caller satisfaction (post-call survey if appropriate)
• Time saved (minutes of human time avoided)

---

A working example

Setup for a small solar business:

• Single inbound number (main line)
• VAPI inbound assistant
• Calendar integration with Cal.com (custom logic via webhook)
• Slack notifications for messages
• Transfer to sales team's mobile during business hours, voicemail/SMS after

Performance after 90 days:

• 250 inbound calls/month
• 60% handled fully by AI (booked appointments, FAQs, hours)
• 30% transferred to humans (sales conversations, support)
• 10% messages taken
• Caller satisfaction: 4.4/5 average (when surveyed)

Saved: ~80 hours/month of receptionist time. Cost: ~$300/month for AI platform + voice fees. Net: $4,000+/month savings.

---

Sources

VAPI inbound configuration from vapi.ai/docs. Industry voicemail leave-rates from Salesforce research and CallTrackingMetrics studies. Receptionist salary ranges from BLS Occupational Outlook Handbook. AI deployment metrics from typical SMB engagements.

Want help deploying inbound AI for your business? Let's talk — typical setup is 1-2 weeks.