Overview
This demo simulates an IVR (Interactive Voice Response) appointment scheduling system — the same type of AI that handles inbound calls for healthcare providers, financial services firms, and enterprise service desks. Instead of navigating touch-tone menus, callers speak naturally and the AI understands intent, gathers details, confirms actions, and updates the calendar in real time.
How It Works
- Click "Start" — The assistant greets you and explains your options.
- Speak or type — Use the microphone button to speak naturally, or type your request in the text input. Both work the same way.
- The AI responds — It asks follow-up questions if needed (purpose, duration, attendee), then confirms the details before acting.
- Watch the calendar update — The moment you confirm, the appointment appears on the live calendar to the right. Reschedule or cancel existing appointments the same way.
Interaction Examples
Try saying:
- "Schedule a one-hour project review with Mark Chen on Wednesday at 10am"
- "Move my client review to Friday afternoon"
- "Cancel the strategy meeting"
- "What meetings do I have this week?"
Technical Notes
This demo runs in your browser using:
- Speech-to-text: Web Speech API (Chrome and Edge — gracefully degrades to text-only in other browsers)
- Text input: Works in all browsers as an alternative or complement to voice
- Conversation AI: GPT-4o-mini maintains context across turns and extracts structured intent
- Text-to-speech: Browser speech synthesis (Microsoft Edge provides the best voice quality using its neural voices)
- Calendar: Live React state with dynamically generated sample appointments — no data is stored or transmitted beyond your browser session
From Demo to Production
This demo simulates a shared calendar with conflict detection and availability checking. A production deployment would extend this significantly:
Calendar & Integration
- Multi-calendar sync — Connect to Google Calendar, Outlook/Exchange, or any CalDAV provider to read and write real appointments across multiple people's calendars
- Cross-calendar availability — Check free/busy status across all participants before booking, not just a single shared calendar
- Room and resource booking — Reserve conference rooms, equipment, or other shared resources alongside the appointment
Automated Workflows
- Confirmation emails — Automatically send calendar invites and confirmation emails to all attendees after booking
- Pre-appointment documentation — Trigger workflows to send intake forms, questionnaires, or preparation materials (e.g., a new patient sending medical history forms before their first visit)
- Reminder sequences — Automated SMS/email reminders at configurable intervals (24 hours, 1 hour before)
- Follow-up actions — Post-appointment surveys, notes distribution, or next-steps emails triggered automatically
- Waitlist management — If a preferred slot is taken, offer to add the caller to a waitlist and notify them if it opens up
Telephony
- Twilio Voice for PSTN connectivity — callers dial a real phone number and the AI handles the conversation
- OpenAI Whisper or Deepgram for higher-accuracy cloud speech-to-text
- OpenAI TTS or ElevenLabs for premium voice quality (this demo already uses OpenAI TTS)
Industry-Specific Examples
- Healthcare: New patient intake workflow — schedule appointment, send registration forms, verify insurance, send pre-visit instructions
- Legal: Client consultation booking — collect case type, send engagement letter, schedule with the appropriate attorney
- Property Management: Maintenance request scheduling — log the issue, check technician availability, dispatch and confirm
- Salon/Spa: Service-specific scheduling — match appointment duration to service type, assign the right stylist/technician
Real-World Challenges
| Challenge | Why It Matters |
|---|---|
| Speech recognition accuracy | Accents, background noise, medical/legal terminology, and phone audio quality all degrade transcription |
| Conversation recovery | When the AI misunderstands, it needs graceful correction — not "Sorry, I didn't understand" |
| Calendar conflict resolution | Double-bookings, buffer time between appointments, and provider preferences require complex logic |
| HIPAA / data compliance | Healthcare scheduling involves PHI — requires encrypted channels and audit trails |
| Caller patience | People hang up if the AI takes more than 2–3 seconds to respond or asks too many questions |
| Handoff to human | Some callers want a human immediately — the system needs to detect frustration and escalate |
Cost Estimates (Monthly)
| Component | Starter | Growth | Enterprise |
|---|---|---|---|
| AI API (conversation) | $30–100 | $100–400 | $400–1,500 |
| Telephony (Twilio Voice) | $50–200 | $200–800 | $800–3,000 |
| Speech-to-text (Whisper/Deepgram) | $20–80 | $80–300 | $300–1,200 |
| Calendar integration | $0–50 | $50–200 | $200–800 |
| TTS (premium voices) | $10–40 | $40–150 | $150–600 |
| Total monthly | ~$100–500 | $500–2,000 | $2,000–8,000 |
ROI Definition
- Primary metric: Call handling capacity — target 60–80% of scheduling calls handled without human agent
- Secondary metrics: After-hours availability (capturing appointments at 9pm that would otherwise be lost), no-show reduction via automated reminders
- Break-even: Typically 2–3 months
- Concrete example: A medical practice with 2 front desk staff spending 60% of their time on phone scheduling at $18/hr = $4,492/month. AI handles 70% = $3,144/month saved vs ~$1,000/month tool cost. After-hours booking alone (capturing 15–20 additional appointments/month at $150 average visit value) adds $2,250–$3,000/month in revenue.
Build It or Buy It?
Telephony is a floor under both routes — what is actually different
Unlike most automation, you cannot build your way out of the largest variable cost. Carrier minutes are charged whoever owns the software. Twilio's US list prices:
| Item | Price |
|---|---|
| Inbound to a local number | $0.0085/min |
| Outbound to a local or toll-free number | $0.0140/min |
| Inbound to a toll-free number | $0.0220/min |
| Browser / app leg | $0.0040/min |
| Local number | $1.15/month |
A four-minute inbound scheduling call on a local number is about 3.4¢ in carrier cost. A thousand such calls is about $34/month — small enough that it is rarely the deciding factor, and worth knowing so it is not mistaken for one.
Speech recognition, text-to-speech and the model sit on top, and are also charged either way. What genuinely differs between build and buy is the scheduling logic and the integrations, not the voice stack.
Where the real work is — and it is not the conversation
The demo shows the easy half. Production difficulty is concentrated in three places, none of them speech:
1. The calendar is the hard system, not the phone. Real availability means provider working hours, appointment-type durations, buffers, room and equipment constraints, double-booking rules, and cancellation windows. Most practice-management and field-service systems expose this through a limited API, or through no API at all — which is the single most common reason these projects stall.
2. Identity before action. Booking, moving or cancelling an appointment requires knowing who is calling with enough confidence to act. Caller ID is a hint, not proof. In healthcare this is a compliance question, not a UX one.
3. Failure has to be graceful in real time. A web form can show an error. A caller on hold cannot be left in silence, and cannot be told to try again later without losing the appointment. Every branch needs a human fallback that actually connects.
Choose to buy when your scheduling system has a vendor-supported voice add-on. It will already know your availability model, which is the expensive part to replicate.
Choose to build when you have several systems to reconcile, unusual scheduling rules, or you are being quoted per-seat pricing for something that should be per-call.
A note on this market: voice scheduling vendors are overwhelmingly sales-led, and published per-minute or per-call pricing is rare — which is exactly why the Twilio numbers above are useful. They give you a floor to sanity-check any quote against. If a vendor quotes several dollars per call, you now know roughly what portion of that is carrier cost and what is margin and software.
What to expect — containment rates and the failure modes that matter
Containment. 60–80% of routine scheduling handled without a human is a realistic target. Routine is doing real work in that sentence: new patient intake, insurance questions and anything emotionally charged should escalate by design, not by accident.
Latency is the quality bar. Callers tolerate a wrong answer better than a two-second silence. Response time budget matters more here than in any chat interface, and it constrains model and TTS choices more than accuracy does.
Where it goes wrong
- Accents, background noise and speakerphone degrade recognition unevenly across your caller base. Test with real recordings, not clean studio audio.
- The caller says "next Tuesday" three days before a holiday. Relative date handling is a persistent source of wrong bookings.
- Someone books successfully, then calls back to change it. Modify-and-cancel flows are consistently harder than booking and are consistently under-built.
- The integration writes the appointment but the confirmation fails, so the caller does not know it worked and books again by phone with a human.
Measure completed appointments, not completed calls. A call that ends politely without a booking is a failure that most dashboards record as a success.
Sourcing note. Twilio Voice figures are US list prices read from Twilio's pricing page in August 2026 and vary by country and number type. Named voice-scheduling vendors are omitted deliberately: their pricing is quote-only, and we would rather give you a floor to negotiate against than print a number we cannot cite.
Use Cases
- Healthcare patient appointment scheduling and rescheduling
- Financial services client meeting booking
- Legal and professional services intake scheduling
- Internal IT service desk appointment routing
- Property management and real estate showing coordination
- Franchise and multi-location appointment coordination
Technology Stack
- Conversation AI: OpenAI GPT-4o-mini (structured intent extraction + multi-turn context)
- Speech-to-Text: Web Speech API (demo) / OpenAI Whisper or Deepgram (production)
- Text-to-Speech: OpenAI TTS (demo + production) / ElevenLabs (premium option)
- Telephony: Twilio Voice (production PSTN connectivity)
- Calendar: React state (demo) / Google Calendar, Outlook, CalDAV (production)
- Frontend: React client with real-time calendar visualization
- Backend: Next.js API route (serverless)
Want This for Your Business?
A production voice scheduling system with real phone number, calendar integration, automated reminders, and HIPAA-compliant data handling typically deploys in 3–5 weeks and starts at $4,000. Multi-location and multi-provider configurations scale from there.
This demo uses GPT-4o-mini for conversation and OpenAI TTS for natural speech. No appointment data is stored — all calendar state resets when the page is refreshed. Voice works in Chrome and Edge; text input works in all browsers.