Clinics & Service Businesses
AI Voice Receptionist and Booking Agent
An agent that answers every call, books into the real calendar, and hands off to a human when it should. How we'd build it, in what order, and where it breaks.
Who it is for
Typically the owner-operator of a one to five location service business (a med spa, a dental practice, a home-services company) who can only guess how many calls were missed, and who may already have tried a generic AI receptionist that mishandled a call or failed to book into the real calendar.
First working version
6 to 10 weeks
A planning estimate, firmed up once we see your real stack.
Foundation we would extend
AI Voice Caller
Our own AI Voice Caller is in active development and is the starting point we'd extend, not a finished product to resell. See our products
The problem
Who has this problem, and why it matters
Small clinics, med spas, home-service companies and wellness businesses lose bookings every week to missed calls, after-hours voicemail and front-desk staff who are also doing five other jobs. Most of these businesses cannot say how many calls they missed last month, only that it was too many.
Hiring more reception staff is expensive, hard to schedule around peak call times, and does not scale across locations. The calls that get missed are often the most valuable ones: a new customer with an urgent need, calling at the one moment they were ready to book.
Market
Where existing tools stop
Voice AI platforms such as Vapi, Bland and Retell made the raw technology easy to access, and a wave of wrapper products launched on top of them. Many stop at 'the AI answers the phone'.
In our read, fewer of them handle the operational parts well: real calendar write access with conflict resolution, multi-location routing, and a clean human handoff with context. The gap is no longer the voice technology. It is the reliability around it.
This is our read of the market, not a formal study.
The solution
What we would build, in order
- 01
Answer and book
A voice agent answers inbound calls, handles common questions and books into the business's real calendar.
- 02
Two-way sync
Appointments, reschedules and cancellations stay accurate in both the booking system or CRM and the agent's view.
- 03
Follow-up
SMS or WhatsApp confirmations, reminders and no-show recovery run automatically after the call.
- 04
Smart handoff
Urgent or unusual calls transfer to a human with a full summary, so the caller never repeats themselves.
- 05
Visibility
A dashboard shows call volume, booking conversion, missed-call recovery and transcripts.
Architecture
How it works, end to end
Caller dials the business number
Routed through a phone provider (Twilio or a SIP trunk).
Streaming voice pipeline
Speech-to-text, the language model and text-to-speech run as a stream so replies feel immediate and callers can interrupt.
Agent with tools
The agent does not guess. It calls tools to check availability, book, reschedule or look up a customer.
Booking system of record
Google Calendar, Calendly or the business's own booking API, written to with conflict checks.
After the call
SMS confirmation, CRM update, transcript and summary stored for the dashboard.
Handoff path
Emergencies and unusual requests transfer to a person with the conversation summary.
Tech stack
What we would use, and why
Phone layer
Twilio or a SIP provider
Number provisioning and porting, call control and transfer.
Voice orchestration
Vapi or a custom streaming pipeline
Controls latency and turn-taking. Buy it first, own it later if cost or control demands.
Reasoning
LLM API with tool calling
Structured actions instead of free-text promises the system cannot keep.
Booking and CRM
Google Calendar, Calendly, HighLevel or Zoho
The calendar is the source of truth. The agent writes to it, never around it.
Data and jobs
Postgres and a job queue
Transcripts, retries and follow-up messages must survive failures.
Observability
Call logs, traces and cost-per-call tracking
You need to know a call went wrong before the customer tells you.
Prototype vs production
A demo is not a product
You can prototype this yourself
- +A scripted greeting and a few FAQ answers on a voice platform such as Vapi
- +A booking link the agent reads out or texts
- +Enough to hear what the experience feels like and decide if it is worth doing properly
We encourage it. A working prototype is the best way to find out what you actually want.
What production needs
Double-booking prevention
Two callers can ask for the same slot at the same moment. Without locking and conflict resolution, the agent books both.
Natural turn-taking and interruptions
Callers talk over the agent, pause mid-sentence and call from noisy places. Latency and barge-in handling decide whether it feels human or frustrating.
Timezone, holiday and multi-location logic
A booking that is correct in one city is wrong in another. Locations have different hours, staff and rules.
Strict escalation for emergencies
Some calls must never be handled by an AI. The rules for detecting and transferring them need review by the business, not just an engineer.
Recording and consent rules
Call recording laws differ by region and the agent must disclose and respect them.
Number porting and carrier setup
Moving a live business number is slow and can interrupt service if done carelessly.
Failure fallbacks
When the AI provider, phone provider or calendar API is down, calls must still reach a human.
Cost control
Per-minute costs add up across telephony, speech and the model. A long, looping call should be cut off, not billed.
Our approach
Where we would start
We would build the booking engine before polishing the voice. A friendly agent that double-books is worse than no agent, so calendar write access and conflict resolution come first, tested against the business's real calendar setup.
In the first week we would confirm the things that cause delays later: how the existing number can be ported or forwarded, which recording-consent rules apply, what permissions the calendar API actually grants, and what counts as an emergency for this business. Handoff rules and multi-location routing are designed in the first phase, not added after launch.
FAQ
Common questions
Can we just build this ourselves with Vapi or a similar platform?+
For a prototype, yes, and we encourage it. The gap shows up in production: calendar conflicts, handoff, failure fallbacks, consent rules and cost control. Those are the parts that decide whether the business can trust it with real customers.
Does this replace the front desk?+
No. It handles routine calls, after-hours calls and overflow, and passes anything urgent or unusual to a person with a summary. The aim is fewer missed calls, not fewer staff.
Is it HIPAA compliant for clinics?+
It can be built that way, but compliance is not automatic. It needs signed agreements with every vendor that touches patient data, encryption, access controls and audit logs. We would scope this in the first week and tell you honestly what it takes.
How long does a first version take?+
Our planning estimate is 6 to 10 weeks for one location, longer for multi-location with custom integrations. We firm that up once we see your booking system.
Sales & Operations
CRM-Grounded Sales and Inquiry Agent
High-value inquiries get slow, inconsistent replies that rarely use the business's real pricing and availability.
Read concept →Enterprise
Multi-Agent Enterprise Assistant on Microsoft 365
Teams want AI that acts on company data without sending it to an uncontrolled tool.
Read concept →Have something like this in mind?
Bring your prototype
A half-formed idea, a working demo or a full spec. We will tell you honestly what is worth building first and what you can safely keep doing yourself. Typical response under 2 hours.