To set up an AI missed-call receptionist you need four components: a way to detect unanswered calls, a WhatsApp Business API number to reply from, an AI conversation flow that qualifies the caller and offers appointment slots, and a connected calendar. A working system takes roughly one to three weeks to build.
Key takeaways
- Four building blocks: call detection → instant WhatsApp reply → AI qualification → calendar booking.
- Detection works via call forwarding or a call-tracking layer.
- The reply must go out in under 60 seconds — this is the entire point of the system.
- Scope the AI narrowly and make human handoff instant and obvious.
- Meta's WhatsApp Business Verification is the slowest step; start it first.
What you will need before you start
| Component | Purpose | Notes |
|---|---|---|
| Call detection method | Knows when a call went unanswered | Forwarding, SIM-based app, or call-tracking service |
| WhatsApp Business API number | Sends the reply | Must be a separate dedicated number |
| Meta Business Manager, verified | Required for API access | Start this first, it is the bottleneck |
| Automation platform | Runs the logic | A BSP dashboard, a workflow tool, or custom code |
| AI model or conversation engine | Understands and replies naturally | Scoped with clear instructions |
| Calendar | Holds real availability | Google Calendar or your practice software |
Step 1: Decide how you will detect a missed call
This is the piece most guides skip, and it determines everything else. Three workable approaches:
Conditional call forwarding. Configure your line so unanswered calls forward to a number connected to your automation. Your carrier supports this. Simple, cheap, and keeps your primary number intact.
A call-tracking layer. Route calls through a tracking number that logs answered versus missed and fires a webhook. Adds analytics as a bonus — useful if you also run ads and want cost per booked appointment reporting.
A device-level app. An Android app on the business phone that watches the call log and triggers a webhook on a missed call. Cheapest, but fragile — it depends on one physical handset staying on and charged. Fine for testing, not for production.
For most businesses, conditional forwarding is the right answer.
Step 2: Get your WhatsApp Business API number live
Start this on day one because Meta's verification takes the longest.
- Create a Meta Business Manager account and complete Business Verification with your registration documents.
- Choose a Business Solution Provider.
- Register a dedicated number — one with no existing WhatsApp account on it.
- Get your display name approved.
- Submit your initial message templates.
The full walkthrough is in our WhatsApp Business API India guide.
Step 3: Write the first message
This single message determines whether the system works. It has four jobs: identify the business, acknowledge the miss, reduce friction, and ask exactly one question.
A good first message:
Hi! This is Dr. Mehta Dental — sorry we missed your call just now. I can help you right here on WhatsApp. What did you need help with?
Why it works: it names the business immediately so the message is not mistaken for spam, apologises without grovelling, offers a channel the customer already uses, and asks an open question that invites a reply.
Common mistakes to avoid:
- Sending a bare link with no context — reads as spam and gets blocked.
- Asking three questions at once — people answer none of them.
- Being formal and corporate — kills the conversational feel that makes WhatsApp work.
- Not identifying the business — the recipient has no idea who is messaging them.
Step 4: Design the qualification flow
Keep it to two or three questions maximum before offering a booking. Every additional question loses people.
Examples by business type:
| Business | Question 1 | Question 2 | Then |
|---|---|---|---|
| Dental clinic | What's the concern? | Is it urgent or routine? | Offer slots |
| Gym | Fitness goal? | Preferred time of day? | Offer trial slots |
| Solar installer | Do you own the roof? | Monthly electricity bill? | Offer survey slots |
| Interior designer | Property type and size? | Budget range? | Offer consultation |
| Broker | Budget and configuration? | Buying timeline? | Offer site visit |
| CA | Salaried, freelance or business? | What service do you need? | Offer consultation |
The qualification is not an interrogation. Each question should feel like a helpful step toward getting the customer what they came for.
Step 5: Connect a real calendar
Offering fake or stale availability destroys trust instantly. The AI must read genuine free slots and write the confirmed booking back.
Requirements:
- Read availability in real time, respecting working hours, breaks and existing appointments.
- Offer two or three specific options, never "when are you free?" — specificity converts far better.
- Write the booking back immediately and confirm in the chat.
- Send the location pin and any preparation instructions.
- Schedule reminders at 24 hours and 2 hours before.
Step 6: Set the guardrails
Write these rules explicitly into the AI's instructions:
- Never give professional advice — clinical, legal, financial or structural.
- Never quote prices outside approved ranges.
- Never claim to be human. Identify as the business's assistant if asked.
- Always allow escalation. "Talk to a person" must work instantly, at any point.
- Flag emergencies to a human immediately rather than booking them.
- Stay in scope. If the question is unrelated to the business, hand off.
Step 7: Test properly before going live
Run at least twenty test conversations covering:
- The straightforward booking
- A caller who asks about price
- A caller who is rude or tests the bot
- A caller who types in Hinglish or a regional language
- A caller with a genuine emergency
- A caller who asks for a human immediately
- A caller who goes silent midway
- A caller wanting a slot when you are closed
Have someone outside the project run these. You know what the system expects; a stranger does not.
Step 8: Measure the right things
| Metric | Target |
|---|---|
| Median first response time | Under 60 seconds |
| Reply rate to the first message | Above 50% |
| Conversations reaching a booking | Above 30% |
| Escalation rate to a human | 15–25% is healthy |
| Block or report rate | Near zero |
A very low escalation rate is a warning sign, not a success — it usually means the AI is answering things it should be handing over.
Frequently asked questions
How long does it take to set up an AI missed-call receptionist?
One to three weeks in most cases. Meta's Business Verification is the longest step and should be started first. Flow design and testing typically take a few days once API access is live.
Can I build this myself?
Yes, if you are comfortable with webhooks, an automation platform and Meta's approval process. The technical work is moderate; the difficulty is in conversation design, guardrails and edge-case handling, which is where most self-built systems fail.
What happens if the AI cannot answer something?
It should escalate to a human immediately and say so plainly. A well-configured system routes anything advisory, unusual or emotionally charged to a person, passing the full conversation history so the customer never repeats themselves.
Will this annoy people who called by mistake?
Rarely, because the message identifies your business, is polite and is easy to ignore. Wrong numbers simply do not reply. Keep the first message to one polite question and never follow up on a non-response more than once.
Or have it built for you
fizaki builds and maintains AI missed-call receptionists for Indian local service businesses — verification, templates, conversation design, calendar integration and guardrails included, live in about two weeks.
Related reading: what an AI receptionist is and missed call statistics for Indian small businesses.
Stop losing leads to a phone nobody answers.
Book a free 20-minute audit — we'll show you exactly how many leads you're leaking and what they're worth.