After-hours AI receptionist: what it should handle and when to hand over

Most small businesses don't need a voice agent during business hours. They need one at 7.40pm when a customer finally gets around to ringing, on Saturday morning when the owner is at a kid's footy game, and at 2.15pm when everyone is on a job. An after-hours AI receptionist answers those calls, deals with the predictable ones and leaves you a tidy list of the rest.

Inbound answering is the lowest-risk place to start with voice AI, for the reasons set out in our guide to AI voice agents in Australia: the caller chose to ring you, so you're not cold-calling anyone, and the cost of the current process, voicemail that nobody returns until tomorrow, is easy to measure.

Decide the job before you choose the voice

Pull four weeks of call logs and listen to a sample of voicemails. Sort the calls into buckets. For most owner-operated businesses, after-hours calls fall into a handful of types, and only some of them should go to an AI.

Our opinion: don't let the agent resolve complaints, even simple ones. An apology from software reads as a brush-off. Take the message, set expectations and let a person call back.

Escalation rules that actually work

The single most important design decision is when the receptionist stops trying and gets a human. Write the triggers down; don't leave them to the model's judgement.

• Keywords and phrases: "emergency", "flooding", "gas", "hurt", "complaint", "lawyer", "cancel my account" and anything else specific to your trade.
• Repeated failure: if the agent has misunderstood twice, offer a person rather than a third attempt.
• Direct request: "Can I speak to someone?" always gets a yes, even if the yes is a call-back time.
• Out-of-scope promises: price commitments, refunds, warranty decisions and anything contractual.

Escalation has to land somewhere real. For urgent calls that means a live transfer or an immediate alert to whoever is on call; for everything else, a task with the caller's name, number, a two-line summary and a promised call-back time. If nobody is rostered to act on urgent alerts after hours, don't offer urgent help. Tell callers honestly what will happen and when.

Booking without double-booking

Letting an AI write to your calendar is where most of the value is and where most of the mess comes from. Start conservatively:

1. Give the agent a defined set of bookable slots or job types, not your whole calendar.
2. Have it offer two or three options rather than asking an open "when suits you?".
3. Read the booking back: day, date, time, address and what the job is.
4. Send an SMS confirmation immediately, with a way to change it.
5. Review every AI-made booking each morning for the first two weeks.

If your work needs a site visit or an assessment before you can quote, book the assessment, not the job. An AI that books a two-hour slot for what turns out to be a full bathroom renovation creates more work than it saves.

The SMS that closes every call

Every call should end with a text the caller can keep. It confirms the agent understood, it gives the customer something to forward to a partner, and it catches errors early because people reply when something is wrong.

That is a factual, transactional message. Keep it that way: no promotion, no "10% off your next service". Adding marketing makes it a commercial message under the Spam Act, with consent and unsubscribe obligations. We cover the distinction, and the new SMS sender ID rules, in SMS follow-up after a call. And if you can't answer at all, missed call text back is the lighter-weight fallback.

A sample script for an AI receptionist

Keep the opening short, honest and complete. It should name the business, say it's an AI, mention recording and offer a person.

Then a short list of questions the agent must get through for a new job: name, best number, suburb and address, what's wrong, how urgent, and preferred times. The recording line isn't decoration; it addresses federal and state recording rules, which we explain in call recording laws in Australia.

Mistakes we see with AI answering

• Pretending to be human. Callers usually work it out, and then they distrust everything else the agent said.
• Over-long greetings. Every extra sentence before the caller can speak increases hang-ups.
• No way out. An agent that loops instead of offering a person turns a missed sale into a complaint.
• Answering from the whole website. Give it an approved FAQ. Websites contain old prices and services you no longer offer.
• Nobody reads the recaps. If call summaries pile up unread, the agent is just a more expensive voicemail.

How the morning hand-over looks in practice

The point of after-hours answering is a clean start to the next day. With Fulcrum Lyre, each call's outcome is logged to the CRM and a recap lands in the chat channel you already use, so the morning list looks like this: three bookings made (check and confirm), one urgent call escalated at 9.52pm (already handled by the on-call plumber), two quote requests (call back before 10am), one complaint (owner to ring). Cortex agents draft any follow-up emails or texts and you approve them before they go out.

Measure it the way you'd measure a person on the phones: how many calls were answered, how many became bookings, how many were escalated and how many callers hung up in the greeting. Our AI voice agent metrics article has a simple dashboard.

When a human answering service is the better choice

AI answering isn't always the right call. If your after-hours calls are mostly distressed people, such as property emergencies involving vulnerable tenants or healthcare calls, a trained human service may serve them better. If you get two after-hours calls a month, a good voicemail greeting and a reliable morning call-back routine costs less. And if your callers skew older or speak mainly in languages your agent hasn't been tested in, test carefully before switching; see latency, accents and voices for how.

If the numbers do stack up, start with one call type and a fortnight of daily review. The Lyre page explains how inbound calls, SMS and CRM logging connect.