Set up an AI phone assistant
Build a phone helper that answers approved questions, handles bookings, and passes uncertain calls to a person.
Original course title & technical overview
AI Voice Receptionists That Book Appointments. Build a phone agent that answers every call for a local business, answers questions from their knowledge base, books into their real calendar, texts confirmations and hands off to a human when it should. Sell it as a monthly service.
A new word? Look it up in plain English →
2026 EDITIONWhat you can offer
Build a phone helper that answers approved questions, handles bookings, and passes uncertain calls to a person.
Detailed deliverables
A 24/7 AI receptionist on the client’s phone number: answers FAQs, qualifies callers, books and reschedules appointments in Google Calendar or their booking system, texts confirmations, takes messages, and transfers urgent calls. Plus a weekly call report.
Tools (free plans have limits): Retell AI or Vapi (voice agent platform), Twilio or the platform’s own numbers, Cloudflare Workers + D1 (A4 stack) for tools and logging, or n8n for no-code flows, Google Calendar API or Cal.com / Calendly / the client’s booking system, Claude for building, testing and the weekly report
You will make
- A voice agent on Retell or Vapi with a persona, a tested conversation flow and a knowledge base built from the client’s facts
- Tools the agent calls mid-conversation: check availability, book, reschedule, cancel, take a message, transfer, all served by a Workers webhook
- Calendar integration that books real events with confirmation texts and a human-handoff path
- A test suite of 20 scripted calls covering names, numbers, accents, interruptions and edge cases, with pass criteria
- A live receptionist for the capstone client with the weekly report and the monthly retainer package
Inside this course
5 lessons · 7 AI instruction templates · 5 worked examples · 3 exam versions · final practice projectLesson 1 is open as a free preview, ready-to-use AI instructions included. Sign up free to unlock the rest and try AI instructions in the practice area.
- 1🔒 Sign inWhat you will learn · 2 min
- 2🔒 Sign inSet up your tools · 10 min
- 3Preview freeA phone assistant listens, decides how to respond, and speaks. It should use approved information, confirm bookings, and hand over when unsure.
- 4🔒 Sign inPersona, prompt and knowledge base · 45 minChoose the assistant’s greeting and style, then build its reference information from facts the client approves. Clearly mark unanswered questions.
- 5🔒 Sign inTools: availability, booking, messages, transfer · 75 minConnect the assistant to the actions it needs, such as checking a calendar or taking a message. Test each action and its confirmation step.
- 6🔒 Sign inTesting calls: names, numbers, accents, chaos · 60 minTry difficult names, dates, interruptions, noise, and unexpected questions before going live. Check that uncertain situations reach a person.
- 7🔒 Sign inGo live, report weekly, and run it as a service · 45 minChoose when the assistant answers and how calls reach a person. Review its work regularly and agree on what ongoing support covers.
- 8🔒 Sign inCertification exam · 25 min10 questions · 80% to pass · three versions, a retake is never the same test.
- 9🔒 Sign inFinal project: A 24/7 receptionist for Glow Aesthetics Med Spa · 75 minA 24/7 receptionist for Glow Aesthetics Med Spa: graded on 5 criteria, pass at 15/20.
- 10🔒 Sign inCheck your work · 15 min
- 11🔒 Sign inFeedback and certificate · 5 min
- 12🔒 Sign inPublish your service · 20 min
Final practice project: A 24/7 receptionist for Glow Aesthetics Med Spa
A practice assignment from Glow Aesthetics Med Spa (Medical spa (injectables, laser, facials)). You will see the scoring guide (rubric) before you start.
| What we check | What earns the full 4 points |
|---|---|
| Brief followed | All five deliverables; consults bookable, treatments routed to consult; after-hours mode planned first |
| Safety | Knowledge-base-only enforced; approved prices only; medical/insurance questions declined to consult; emergency path; read-back before booking; disclosure line; no card data by voice |
| Engineering | Guardrails in Worker code; secret header verified; idempotent booking proven; SMS sent; D1 logging; refusal messages speakable |
| Conversation quality | Suite passes on hard names, numbers with corrections, relative dates, interruptions and off-script questions; latency under 1.2 s |
| Service operations | Go-live plan executable by the phone provider; handoff rules clear; weekly report leads with recovered calls and bookings; retainer cap and exclusions stated |
Honest expectations. Voice agents fail in specific, predictable ways: interrupting, mishearing names and numbers, booking the wrong day, and looping. This track tests every one of those on purpose before a real caller hears the agent. Never let an agent give medical, legal or pricing commitments the client did not approve in writing; the knowledge base is the only thing it may say.