# AI phone receptionist: coming soon to Pilotiko

> Pilotiko's AI receptionist: what is built, what has not been tested on a real phone line yet, and how to cover your calls meanwhile. Join the waiting list.

Updated on October 15, 2026 · https://app.pilotiko.com/en/product/ai-receptionist

Pilotiko's AI receptionist is coming soon: it does not answer any real calls yet. The code exists (conversation, booking, hand-off to a person), but it has only run in simulation, never on a real phone number. It is not included in any plan. You can join the waiting list by booking a demo.

## In short
- Status: coming soon. Pilotiko has not answered a single real call.
- Built and tested in simulation: French and English conversation, open-slot lookup, booking with spoken confirmation, transfer to a person.
- Still missing: a verified live number, French voicemail transcription, website chat for visitors.
- Available today: online booking from real open slots.

## What does an AI phone receptionist actually do?

It picks up for you, understands what the caller says out loud, answers simple questions and takes one action: leave a message, offer a time slot, or pass the call to a person.

Three parts work together. A telephony provider receives the call on your number. With Twilio, for instance, the call triggers a web request to your software, which replies with TwiML instructions such as Say, Gather (listen), Dial (connect someone else) or Record [1]. An understanding layer turns the caller's sentence into an intent. Then the software looks up your real data: hours, services, calendar.

What matters most is not the voice but what the receptionist is allowed to do. Booking, cancelling or refunding affects a customer, so those actions need limits. OWASP lists excessive agency as a risk of language-model systems and recommends human approval for high-impact actions [3].

## How far along is Pilotiko's receptionist?

The code is written and tested, but only against simulated Twilio requests and calls played in the browser. No test touches a real phone network, which is why this page says coming soon rather than available.

| Piece | Status | What it means |
|---|---|---|
| Phone conversation (fr-CA, en-CA) | Built, tested in simulation | Never run on a real call |
| Hours, services, open slots | Built, tested in simulation | Reads the same data as online booking |
| Book, cancel, move an appointment | Built, tested in simulation | Requires the caller to confirm out loud |
| Transfer to a person | Built, tested in simulation | During business hours by default |
| Voicemail | Built, tested in simulation | 2 minutes maximum; transcription in English only |
| Automatic text replies | Missing | Texts are logged; a reply waits for a person |
| Website chat for visitors | Missing | No public chat widget |
| Live number connected and verified | Missing | The condition for moving to available |

> The AI receptionist is not offered today and is not part of any plan. Its status will only change after a successful test on a real phone number. No date has been set.

## What does the code already handle in simulation?

The conversation follows set steps: ask for information, pick a service, a date and a slot, then collect name, phone and address before confirming. It also covers cancelling, rescheduling and taking a message. To look up or change an existing appointment, callers first confirm who they are with their last name and civic number or postal code.

By default, understanding runs on fixed rules with no language model. When a model is configured, it only returns an intent and fields, each one validated before use. If speech recognition is unsure, the assistant asks again; after two silences it ends the call politely.

An emergency, a complaint, a refund request or a legal question triggers a transfer to your backup number rather than an improvised answer. When the call ends, an inbound request is created with a factual summary of the reason, topics and next step. Call transcripts are kept for 90 days, then deleted.

## How can you cover calls today without Pilotiko?

To size the issue for your business, the [missed-call estimator](https://app.pilotiko.com/en/tools/missed-call-cost) and the [24/7 coverage calculator](https://app.pilotiko.com/en/tools/staff-a-position-24-7) work from your own numbers.

- **Forward calls to a cell phone**: cheap, but whoever is on call answers in the middle of a job.
- **Voicemail with a promised callback**: simple; state a realistic callback time and keep it.
- **Live answering service**: people answer in your name, usually billed per minute or per call.
- **Booking link in your greeting and on your website**: many calls are only about getting an appointment. [Pilotiko online booking](https://app.pilotiko.com/en/product/booking-and-scheduling), which is available, shows customers your real open slots.

## How many calls would a receptionist need to absorb?

**Example: a Halifax plumbing company taking 60 calls a week**

Hypothetical figures, to show how to count before choosing a solution.

- Calls per week : 60
- Calls outside business hours (20%) : 12
- Calls in business hours with nobody free (15% of the other 48) : 7
- Calls only about booking (40% of the 19) : 8
- **= 19 calls a week without an immediate answer, 8 of which online booking could already handle today.**

## Which privacy rules apply to an AI receptionist in Quebec?

A phone call carries personal information: name, number, address, sometimes a health or housing issue. Quebec's private-sector privacy act says employees or agents may access personal information only if they need it to perform their duties (s. 20) [2]. When a decision is based exclusively on automated processing, the person must be told and must be able to submit observations to a staff member who can review it (s. 12.1) [2].

That is one more reason Pilotiko's receptionist is designed to transfer to a person instead of deciding alone.

## What will Pilotiko build when the receptionist is ready?

Commissioning will build on what exists: connect your Twilio account, place a proof call, check the routing, then get your confirmation before your line is redirected. Actions that touch a customer stay bounded, as with [AI agents](https://app.pilotiko.com/en/product/ai-agents). Confirmations and reminders after a booking depend on [sending being commissioned](https://app.pilotiko.com/en/product/confirmations-and-reminders).

No date is promised. The [roadmap](https://app.pilotiko.com/en/product/roadmap) lists everything marked coming soon, and the [product overview](https://app.pilotiko.com/en/product) shows what you can use today.

## Frequently asked questions

### Does Pilotiko's AI receptionist answer the phone yet?

No. The code has run in simulation and against simulated Twilio requests, never on a live line. Until a real test succeeds, the capability stays coming soon.

### Is it included in a Pilotiko plan?

No. A coming-soon capability is not included in any plan. You can join the waiting list by booking a demo.

### Will it be able to book appointments on its own?

The code covers booking, cancelling and rescheduling, each with the caller's spoken confirmation. It uses the same open slots as online booking, which is already available.

### What happens if the caller wants a person?

The engine transfers to your backup number, during business hours by default. Outside those hours, the caller can leave a voicemail of up to 2 minutes.

### Will it understand Quebec French?

The conversation is written for Canadian French and Canadian English. French speech recognition has never been exercised on a real call, though, and voicemail transcription is only requested in English for now.

### What can I use right now to reduce missed calls?

Pilotiko online booking is available: customers pick a real open slot without phoning. Add the link to your voicemail greeting and your website.

## Sources

1. Twilio Docs, « TwiML for Programmable Voice », https://www.twilio.com/docs/voice/twiml (accessed on 2026-10-15)
2. LégisQuébec, « Act respecting the protection of personal information in the private sector (CQLR, c. P-39.1) », https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1 (accessed on 2026-10-15)
3. OWASP Gen AI Security Project, « LLM06:2025 Excessive Agency », https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ (accessed on 2026-10-15)