Missed calls: voicemail, an answering service or an AI receptionist?

The short answer: count the calls you miss before you buy anything

An AI receptionist answers your business number, talks with the caller, takes their details and sends you a summary. It can help when you miss a steady number of calls you'd want to win, and nobody can answer them. It isn't the only fix, and it isn't always the best one.

Start by counting what you actually miss for two weeks. Then compare four options: voicemail with a fast callback, a call roster, a human answering service and an AI receptionist. If AI wins, keep its job narrow. It should tell callers in the first few seconds that they're speaking with an AI, collect a clean message and pass anything that needs judgement to a person.

Count what you're actually missing

Most owners know they miss calls. Few know how many, when, or what happens next. Your phone or phone system's call log already holds most of the answer. For two normal weeks, write down each missed call to the business number in a simple sheet.

Two weeks of this tells you whether the problem is lunchtime, the hours you're on site, after hours or weekends. It also shows whether your current callback is fast enough. If most missed callers leave a message and you call them back within the hour, a new tool may add little. If many hang up without a message and never call again, you've found the gap worth fixing.

For each missed call, record:

• The day and time, and whether it was during business hours.
• Whether the caller left a voicemail or sent a text.
• How long it took someone to call them back.
• Whether the call was a new enquiry, an existing customer, a supplier or a sales call.
• Whether the new enquiries turned into a quote or a booking.

Four ways to handle the calls you can't take

Each option suits a different pattern of missed calls. Compare them against your two-week count, not against a sales demo.

• Voicemail with a fast callback. It's cheap and simple, and it works when missed callers leave a message and someone calls back quickly. A caller who hangs up without a message is lost, so it only works if your count shows most callers leave one. A greeting that says when you'll call back helps.
• A call roster or divert. Calls go to whoever isn't on the tools, or to a second phone at set times. You keep a person on every call, but only if someone is genuinely free to answer.
• A human answering service. A person answers in your business name, takes a message and can follow a script you write. Callers get a human, but the operator only knows what you've told them, and coverage hours and costs vary between providers.
• An AI receptionist. Software answers every call at any hour, can answer set questions and takes a structured message. It's only as good as its instructions and setup, it speaks for your business, and someone has to check what it's telling people.

What an AI receptionist should and shouldn't do on a call

Whatever it says on the phone, your business says. Treasury's review of AI and the Australian Consumer Law points out that misleading conduct can happen without fault: "a person who acts honestly and reasonably may contravene the prohibition if their conduct is misleading." It adds that "the onus is on businesses to ensure the technologies they use are fit-for-purpose," and that businesses should make sure "the limitations of this technology are made clear to consumers."

That's why the safest setup is narrow. A phone call is harder to correct than a chat window, because the caller hangs up with whatever they heard. Let it answer approved facts, such as your service area, hours and how quotes work. Have it collect the job details and pass anything involving a price, a promised time, a complaint or a refund to a person.

Phone calls also bring situations a website chat rarely does. Decide in advance what it does when a caller describes something urgent or unsafe, asks for a person, or calls about an existing job. A good default is to stop the script, take a callback number and alert someone straight away.

Some AI receptionists can also send the caller a text after the call. A confirmation of the message they left is one thing; adding them to promotions is another, so check the rules before switching that on.

Keep readingSet what an AI can answer and what it must hand over

Keep readingCheck the rules before it texts your callers

Say it's AI and explain the call in the first few seconds

Callers can't see a label, so the greeting has to do the work. The National AI Centre's guidance says "people should know when they're interacting with AI," and calls disclosure especially important where AI "can be mistaken for a human being." The OAIC recommends that public-facing AI tools are "clearly identified as such" to customers.

An AI receptionist also collects personal information: names, numbers, addresses and what's wrong. For businesses covered by the Privacy Act, APP 5 requires reasonable steps to tell people about the collection, such as who you are, why you're collecting it and who it usually goes to. The OAIC's guidelines give a telephone example: explain those matters "at the commencement of the call (perhaps following a template script or using an automated message)."

Check whether the service records or transcribes calls to write its summaries. If it does, treat it like any recorded call and say so up front. A greeting along these lines covers the basics: "Thanks for calling [business name]. I'm an AI assistant taking messages for the team. This call is recorded and transcribed so we can call you back. You can ask for a person at any time." Then link your privacy policy from your website and describe the AI service there.

Not every small business is covered. The OAIC says most businesses with an annual turnover of $3 million or less aren't, but some are regardless of turnover, including health service providers. The greeting is still worth doing either way, so callers aren't left thinking they spoke to a person.

Keep readingCheck the rules on recording and transcribing calls

Questions to ask the supplier before you sign

The OAIC's guidance on commercially available AI products says privacy obligations apply to personal information put into an AI system and to what it generates. It asks businesses to check "who will have access to personal information input or generated" and whether the supplier's terms let it use that data to train its products. Get these answers in writing:

• Where are recordings and transcripts stored, in which countries, and for how long?
• Does the supplier, or any company it uses, train AI models on our calls? Can we turn that off?
• Can it transfer a caller to a person, and what happens if nobody answers?
• How are messages delivered to us, and what happens if the service is down?
• Can we read every call transcript and change the instructions ourselves?
• How is it priced: per minute, per call or a monthly plan? What happens when we go over?
• Can we keep our existing business number, and how do we leave?

Test it with real calls before customers hear it

The OAIC says customer-facing AI should be tested before it goes live: "Before deploying AI products, particularly customer-facing products such as chatbots, you should carefully test them to understand the risks of inaccurate or biased answers." Do it by phone, not just in the supplier's dashboard, and include background noise from a car or a job site.

Write ten test calls based on your two-week count, plus the ones you'd least like it to get wrong. Read every transcript and summary, and check the message reached the right person. Fix the instructions and run them again. Include calls like these:

• A new customer with a straightforward job who just wants a callback.
• A caller who asks, "How much will it cost?" It should explain how quotes work and take details, not name a price.
• A caller who asks, "Can you come today?" It shouldn't promise a time it can't see.
• An existing customer who's unhappy with a recent job.
• A caller describing something urgent or unsafe.
• A caller who says, "I want to speak to a real person."
• A caller with a strong accent, or a bad line, giving a street address and a mobile number.

Worked example: a plumber's lunchtime and after-hours calls

This is a made-up example to show the process, not a client result. A two-person plumbing business in Geelong counts its missed calls for two weeks. It misses 38 calls to the business number. Fourteen are suppliers, existing customers or sales calls. Of the 24 new enquiries, 10 left a voicemail and were called back within two hours; 14 didn't leave a message.

The owner tries the cheapest fix first. The office manager diverts calls to her mobile during lunch, and the voicemail greeting now promises a callback within an hour. The next two weeks still show most unanswered new callers ringing after 5pm, and few of them leave a message.

So the business trials an AI receptionist for after-hours calls only. It says it's an AI in the first sentence, takes the caller's name, suburb, number and problem, gives no prices, and sends a text summary to the on-call plumber. Urgent leaks get an immediate alert. For the first month the owner reads every transcript each morning, and fixes one wrong answer about the suburbs it covers.

Keep readingMake sure every message reaches someone who calls back

Check after a month whether it's earning its place

Run the same missed-call count again and compare. The question isn't how many calls the AI answered; it's whether more of the enquiries you wanted reached a person who called them back. If the summaries are unreliable, or checking them takes longer than calling people back, a better voicemail and callback routine may be the better answer.

Each month, check:

• How many new enquiries reached a person, and how quickly, compared with your first count.
• A sample of transcripts: whether the answers were right and the summaries matched the call.
• Every call it handed over, and whether it reached the right person in time.
• Callers who asked for a person or hung up early, and why.
• What it cost for the month, against the quotes and bookings it helped bring in.