Picture a physio clinic with two treatment rooms. The owner has one question: "Can a bot answer my phone while I'm working on someone's knee?"
Short answer: yes. Longer answer: the bot is the easy part. The hard part is making the website chat, the phone, and the diary all agree with each other.
So this post is the build log version. No magic, just the parts, how they connect, and what to configure so nothing breaks on a busy Monday. I'll use Owlixi as the reference stack, because it ships all the pieces together: an AI Chatbot, an AI Receptionist for calls, browser web calls, a Booking Terminal for staff, and Calendly sync.
The architecture in one paragraph
Think of it as one data source with several clients.
- Source of truth: a single knowledge base. Services, prices, hours, policies, FAQs.
- Inputs: the chat widget, your business phone number, and a web call button.
- Output: the Booking Terminal (plus notifications), or your Calendly calendar.
Every input reads from the same knowledge base. That's the whole design trick. If the chat says the clinic opens at 8 and the voice agent says 9, patients stop believing either one. One source removes that bug class entirely.
The rollout order is just as plain: connect your website or phone line, let the AI train on your knowledge base, then go live with 24/7 call answering, lead capture and booking.
Loading the knowledge base
Garbage in, confident garbage out. This step deserves more time than it usually gets.
Supported sources:
- your website, crawled from its sitemap
- PDF, DOCX, TXT and MD documents
- images via OCR (great for that laminated price sheet someone photographed)
- FAQs you write by hand
- internal team repositories
- audio transcripts
For corrections there's the Training Data Editor. When you need a fixed, word for word answer, use Custom Scripts and Intents.
A typical example: a clinic site that still lists a sports massage the clinic stopped offering a year ago. The AI would have happily sold it. Audit the source before you blame the model.
One thing to know on the free plan: website parsing has a one hour cooldown, so edits take a bit to show up. The paid chat plan removes it.
Input #1: the chat widget
The AI Chatbot is an embeddable widget trained on your site and documents. It answers questions, captures leads and books appointments around the clock, in 95+ languages.
Install options, pick whichever matches your stack:
- one snippet pasted into your site
- the WordPress plugin
- the Shopify app
- the Wix app
The interesting bit isn't the answering, it's the write path. The widget has booking action buttons and is wired to the Booking Terminal. So when a patient types "Any slots Thursday afternoon?", the result is a real booking in the staff schedule, not an email someone has to copy over later.
Patient: Do you treat shoulder injuries, and how long is a first session?
Chatbot: Answers from the clinic's service pages.
Patient: Book me Thursday after 3.
Chatbot: Shows open slots and confirms one.
Input #2: the phone line
Plenty of people will never open a chat bubble. They call. And in a clinic or a barbershop, calls come in exactly when your hands are busy.
The AI Receptionist is a voice agent that answers the phone, handles questions about your services and books appointments in plain conversation.
Routing calls to it
Two options:
- Dedicated business number. Local numbers are available in the US, the UK and Canada.
- Call forwarding. Keep the number your customers already know and forward it to the AI Receptionist. You set this up in the dashboard settings.
Voice config
The speech side uses low latency text to speech with natural pauses, inflections and pitch changes. Speech recognition is tuned to ignore background noise, which is useful in a barbershop with clippers and music running.
Settings you can tune:
- voice selection
- custom greeting
- speaking pace
- multilingual support with automatic locality detection
The human fallback
Every automated front desk needs an exit to a real person. When a caller asks for an operator, a manager or "a real person", the AI asks a short clarifying question and then connects them to your team.
Smart Operator Handoff can ring up to 3 operator lines, either simultaneously or in sequence.
Now the part I'd call the best engineering decision in the product: Mode 2, Protected Mode.
The classic failure goes like this. The AI forwards a call to your mobile, you don't pick up, and the caller lands in your personal voicemail. Awkward for everyone.
Protected Mode adds a check. The AI whispers to you first: "You have a call, press 1 to accept." The caller only gets connected after a human presses 1. If voicemail answers, the call is disconnected automatically.
If you build anything that hands calls to humans, steal this pattern. Verify a person is actually there before you bridge the call.
Input #3: web calls in the browser
There's a middle group of customers. They're happy to talk, they're already on your website, but dialling a number feels like one step too far. Maybe they're on a laptop. Maybe they're abroad.
AI Receptionist Web Calls cover them. A visitor taps a button on your site and talks to the AI Receptionist right in the browser, through their own microphone. Under the hood it's WebRTC, the same tech browsers use for video calls.
What that gives you:
- no phone number needed, tap and talk
- callers can reach you from anywhere in the world, with no phone fees on their side
- a live conversation: questions answered, bookings taken, handoff to your team when needed
- call logs and insights saved for you
The web call widget sits next to the chatbot and installs the same ways: one snippet, the WordPress plugin, or the Shopify and Wix apps.
Architecturally it's just another input. Same knowledge base, same booking destination, same handoff rules. Nothing new to keep in sync.
The output: configuring the Booking Terminal
All three inputs write to one place. The Booking Terminal is the staff screen that receives bookings and cancellations from the chatbot, its action buttons, or the AI Receptionist. Cancellations are processed automatically.
Runs anywhere. It's browser based, so a phone, tablet, laptop or desktop all work. No hardware to buy. Many shops just keep a tablet on the front desk with it open.
Config that matters. These settings are what keep the AI from promising slots your team can't honour:
| Setting | What it prevents |
|---|---|
| Working hours | bookings before opening or after closing |
| Breaks | someone booked into lunch |
| Slot buffers | back to back sessions with no time for notes or cleanup |
| Parallel booking limits | more clients than you have chairs or rooms |
A barbershop with four chairs and a clinic with two rooms need very different parallel limits. Set them on day one, not after the first double booking.
Staff access by QR. From the dashboard you generate QR codes or links with temporary or permanent access tokens. A barber scans it and the schedule opens on their own phone. Temporary tokens are handy for a locum or a weekend cover.
Notifications. New bookings can fire real-time alerts. Email alerts are on the free plan, WhatsApp and Telegram alerts are on paid plans.
Alternative output: Calendly
If your diary already lives in Calendly, you don't need a second calendar.
Connect it with your Calendly Personal Access Token. After that, the AI Receptionist checks availability in real time and books straight into your Calendly calendar.
So the output is configurable: the built in Booking Terminal, or Calendly. Calendly booking comes with AI Receptionist Pro and Ultra.
Observability: logs and transcripts
Treat the front desk like any production system. Ship it, then watch it.
For calls you get real-time transcripts and audio recordings, with structured insights saved automatically. AI Receptionist Pro and Ultra add live call logs and monitoring. On the chat side, the paid plan adds a conversation sessions log and analytics.
My suggested routine: every Friday, read ten transcripts. You'll usually spot a question the AI couldn't answer, a phrase customers use that isn't in your docs, and a moment where a human should have stepped in earlier. Patch the knowledge base or your scripts. That loop is where the quality really comes from.
Pre-launch test plan
Before flipping it on, run this like a QA checklist:
- Ask the chatbot your ten most common questions. Trace wrong answers back to the source page.
- Configure hours, breaks, buffers and parallel limits.
- Call in, ask for a human, and confirm the right phones ring. Let one go unanswered and check Protected Mode keeps the caller out of voicemail.
- Call from a different phone, ideally somewhere noisy, and listen to the greeting and voice.
- Start a web call on desktop and on mobile. Accept the microphone prompt like a customer would.
- Make a booking, then cancel it. Check both in the Booking Terminal or Calendly.
- Send QR access to each staff member and confirm it opens on their device.
- Decide who gets notifications, and where.
- Put the first transcript review in the diary for the end of week one.
Pricing, and the part you actually want
You can test the whole chat to booking flow for free. The Starter plan includes the AI Chatbot, Booking Terminal access and email notifications.
For the voice side:
- AI Receptionist Lite: $29/month after a 3-day trial. Business number, call answering, call forwarding to your team.
- AI Receptionist Pro: $59/month. Adds AI Receptionist Web Calls, live operator handoff, Calendly booking, team sharing, WhatsApp and Telegram alerts, and live call logs.
- AI Receptionist Ultra: $99/month. Everything in Pro, plus unlimited AI calls, 10 premium voices and 24/7 onboarding support.
50% off with OWLIXISALE
Use promo code OWLIXISALE at checkout and pay half.
- AI Receptionist Lite: $29 to $14.50/month
- AI Receptionist Pro: $59 to $29.50/month
- AI Receptionist Ultra: $99 to $49.50/month
Set up the free Starter or grab a plan at owlixi.com.
Back to that physio clinic. A patient taps the web call button while the owner is mid-session, books Thursday at 4, and he only finds out from the alert. That's the whole point: the front desk keeps working while you do the actual work.
Top comments (0)