AI Phone Receptionist

The problem

A missed call is usually a missed booking. Most small teams cannot have someone on the phone every hour the business is open, and nobody is there after closing. Voicemail gets ignored, and callers ring the next business on the list.

Generic AI phone bots fix the coverage problem and create a worse one: they sound confident while getting your prices, hours or policies wrong. For a business, a wrong answer is worse than no answer.

How it stays accurate

This is the part I care about most, and it is designed in from the start rather than added with a prompt.

• Facts come from your material, never from the model. Every business question goes to your documents or your business-info sheet first. The agent may only repeat what those return.
• Exact numbers are read, not searched. Prices, hours, holidays and transfer numbers live in a Google Sheet you edit. They are returned word for word, so the agent cannot quote last year’s price or another service’s.
• “I’m not sure” is a valid answer. Each document answer carries a relevance score. Below a threshold tuned on your own questions, the agent asks one clarifying question, then transfers or takes a message.
• Some topics always go to a person. Refunds, complaints, disputes or anything you mark as sensitive are handed over straight away.
• Gaps become a to-do list. Every question it could not answer appears in a weekly report, so your documents get better over time.

Acceptance before go-live. Zero invented prices or policies, at least 95% correct answers on in-scope questions, 100% correct transfers on explicit and urgent requests, and every test booking correct in the calendar.

How a call flows

flowchart TD
  A([Caller rings your number]) --> B[AI receptionist greets,<br/>says it is an AI assistant]
  B --> C{What does<br/>the caller need?}
  C -->|A question| D[Answer from your documents<br/>or business-info sheet]
  C -->|An appointment| E[Offer free slots,<br/>confirm details, book]
  C -->|A person, or urgent| F[Warm transfer<br/>with a one-line summary]
  D -->|Not confident| F
  F -->|No answer or after hours| G[Take a message,<br/>alert staff]
  D --> H[(Call log, Sheet<br/>and summary email)]
  E --> H
  F --> H
  G --> H

You own everything

The phone number, voice platform, knowledge base, calendar, database and hosting all run in accounts in your business’s name. I build and configure them; you keep full access and can change providers later.

Running costs

All pay-as-you-go, billed by each provider directly to your accounts:

Item	What drives the cost
Voice platform (Retell)	Minutes of calls, plus the phone number
Knowledge base (Pinecone)	Storage and searches — very small for one business
Embeddings and reranking	Per document re-index and per question — small
Hosting for the API	A small container that can scale down when idle
Database (PostgreSQL)	A small managed instance
Cal.com, Google Workspace	Plans you may already have

On a call we estimate the monthly figure from your call volume and typical call length, using each provider’s current prices.

Common questions

Will callers know they are talking to an AI? Yes. The greeting says so, and that the call may be recorded. Being upfront builds more trust than pretending.

What if my prices or opening hours change? Edit the Google Sheet. The change is live without touching the agent. For documents, drop the new version in the shared Drive folder; it is re-indexed and tested before it goes live.

Can it take payments? No, by design. Card details always go to your staff.

What happens when it gets something wrong? Every call is logged with a transcript. Mistakes are found in the weekly review and fixed in the documents, the rules or the thresholds, then re-tested.

Do my staff need to learn new tools? No. Bookings appear in the Google Calendar they already use, and call summaries arrive by email and in a shared Sheet.