Design a small-business AI receptionist for calls, messages, hours, appointment routing and human handoff with Fish Audio speech-layer limits.
An AI receptionist for a small business answers the phone when the owner is working, driving or closed. It can read hours and location, take a structured message, book an appointment through your calendar and route the caller to a person. Fish Audio supplies transcription, streaming text to speech, a consistent voice and API access. Your phone system, calendar, CRM and inbox own routing, availability and follow-up. Fish Audio is not a phone carrier, PBX or answering service.
The best first version is deliberately small: one number, a short menu, a message record and a reliable handoff.
A safe first scope:
Keep these with a person:
For lead qualification before a sales call, see AI voice agents for lead qualification. For booking mechanics, see AI voice agents for appointment booking.
Agent: Thanks for calling Fieldnote Studio. Is this about a new project, an existing one or a invoice question?
Caller: A new project. I would like to book a consultation.
Agent: I have Tuesday at 10 a.m. or Thursday at 2 p.m. with Sam. Which works?
Caller: Tuesday.
Agent: I will book 10 a.m. Tuesday and send a confirmation. Can I confirm your name and email?
Caller: Dana Ruiz, dana@example.com.
Agent: Booked. I am also sending a short intake form. If you need to move it, reply to the confirmation.
The agent books only after the calendar API confirms and sends the record to the owner.
Keep credentials server-side. The agent should call narrow endpoints, not hold a general database connection or an owner password.
A message is not a transcript dump. Capture the caller's name, number, reason, preferred contact time and any deadline, then deliver it in a consistent format:
New call: Dana Ruiz, 555-0142. Wants a kitchen quote before Friday. Prefers a callback after 4 p.m.
Read the message back to the caller and tell them when to expect a response. If the business cannot respond within that window, say so before ending the call.
Disclose recording before collecting personal information and follow the recording rules that apply in your jurisdiction. Collect only what the follow-up requires. Do not ask for card numbers, passwords or government identifiers on a routine reception call.
Keep personal data out of analytics events and debug logs. Define how long transcripts and recordings are retained, and give the owner a way to delete a message or recording.
| Failure | Safe fallback |
|---|---|
| Calendar API times out | Take a message and promise a callback |
| Caller wants a person | Transfer or schedule a callback |
| Request is outside the script | Take a message; do not improvise |
| Caller is hard to understand | Confirm once, then offer a callback |
| Transfer fails | Read the direct number and log the request |
| After-hours emergency | Follow the approved emergency script |
| Owner inbox is unreachable | Use the secondary notification channel |
Log the routing and delivery result so the owner can see whether the message actually arrived.
| Metric | What it tells you |
|---|---|
| Calls answered | How much coverage the agent adds |
| Messages delivered | Whether the handoff works |
| Appointment conversion | Whether calls become bookings |
| After-hours capture rate | Whether missed calls are recovered |
| Callback completion rate | Whether the owner follows up |
| Transfer rate | Whether the scope is narrow enough |
| p95 time to first audio | Whether the call feels natural |
Review the transcripts that ended in a message. They show what the agent could learn to answer next.
Natural speech is roughly 150 words per minute, close to 900 characters. One Fish Audio credit covers about 1,000 characters, so agent speech starts near one credit per minute before repetitions or longer messages.
| Pack | Price | Credits | Approximate agent speech |
|---|---|---|---|
| Starter | $9 | 1,000 | About 16 hours |
| Creator | $29 | 4,000 | About 66 hours, plus commercial rights |
| Studio | $99 | 15,000 | About 250 hours, plus API access |
Phone service, the number, CRM or inbox, calendar, transcription and human time remain separate. The TTS API pricing guide converts characters into a full budget.
It can answer routine calls, read approved information, take messages and route to a person. Fish Audio supplies the speech layer; telephony, call routing, message delivery, CRM and escalation policy remain in your phone and business systems.
Hours, location, basic FAQs, appointment requests, message capture and an expectation for the next business day. True emergencies should follow a separate approved script or route to voicemail.
The agent can classify the request and ask your phone system to transfer to a number or queue. The actual routing, ring groups and voicemail belong to your telephony provider.
Fish Audio supports 30+ languages and voice selection. Test the languages you actually receive calls in, including names, addresses and product terms, before enabling them.
At roughly 150 words per minute, one minute of agent speech is about one Fish Audio credit. Phone service, numbers, CRM, booking and human time are separate costs.