An AI phone answerer takes a message, screens for urgency, and books into a calendar on the plans that support it. It will not price a job and it will not decide who gets sent. NZ vendors list between NZ$99 and NZ$399 a month on their own pages as at 8 September 2026. Run the missed-call maths with your own numbers before you buy.
What it does, on a good day
The thing answering the phone in 2026 is a speech model wired to a phone number, a knowledge base and, if you let it, your calendar. It picks up on the first ring at 7pm on a Sunday, holds a normal conversation, works out who is calling and what they want, and sends you the result as a text or an email.
Four jobs it does reliably. It takes a message with the name, the number, the address and the reason for the call, which beats a voicemail nobody plays back. It screens for urgency, so a burst pipe gets flagged differently from a request for a quote on a bathroom in March. It answers questions off a knowledge base you write: hours, service area, whether you do gas fitting, how long a callout usually takes. And it books, where the plan supports booking.
Booking is the one worth understanding. It works because your calendar has one correct answer: the 10am slot on Thursday is free or it is taken. The agent reads live availability and writes the appointment while the caller is still on the line. Talkify's NZ page says its agent does exactly that against Google Calendar, Microsoft Outlook and Calendly, and Hey Jodie's NZ pricing page says booking and transfer arrive on its higher plans.
What it will not do, whatever the demo shows
Quoting is out. A price depends on what the job turns out to be once someone has stood in the room, and the caller's description is the least reliable input in the whole process. Any phone agent that gives a number over the phone is giving a number you will have to walk back. The most it should say is a callout fee you have published, word for word, and even that wants care.
Dispatch is out. Deciding which van goes where, in what order, with which materials on board, is a scheduling decision with your margin sitting on it. A phone agent cannot see who is running late or whose ticket covers the work. Let it capture the job. Let a person allocate it.
Anything with money attached is out. Taking a deposit, agreeing a payment plan, confirming a variation. If a caller pushes on price, the correct behaviour is a transfer or a promised callback, and you want that path tested before you go live.
Products with NZ availability, and what they list
Three vendors selling into New Zealand, figures taken from their own pages on 8 September 2026. I have not run any in production, so this is a read of what each publishes rather than a review.
| Product | Listed price | What the page says is included |
|---|---|---|
| Hey Jodie | NZ$99 Basic, NZ$199 Professional, NZ$399 Premium a month | Unlimited minutes on all tiers. Urgency detection, job captured and texted to you. Booking and call transfer on the higher plans. States it presents as an AI rather than a named human. |
| Talkify | $129 a month founding rate, $199 standard | 200 minutes included, overage from 10c a minute. Up to 3 agents on one number, 3 concurrent calls. Books against Google Calendar, Outlook and Calendly. Warm transfer to your mobile on urgent calls. No setup fee, no lock-in. |
| AppAI | Flat $350 a month | No minute caps, no lock-in. Answers 24/7 in a New Zealand voice, works out who rang and why, and texts you an SMS summary. |
The NZ market for an AI answering service sits between roughly NZ$1,200 and NZ$4,800 a year, on the three vendors' own pages as at 8 September 2026. That is the band to hold in your head. Prices move and founding rates expire, so read the vendor page before you budget.
Check whether minutes are capped on any vendor, because a per-minute overage turns a busy month into a bill you did not plan.
The missed-call maths, with the caveats attached
The reason to buy one is calls you are not answering. The reason to be careful is that the industry sells that idea with statistics nobody sources. I built the missed-call cost calculator so you can do the sum with your own numbers and see every assumption labelled, which is the only version of this maths I will stand behind.
The formula is four inputs multiplied together. Calls missed a week, times the share that were someone new, times your close rate, times your average job value. Worked through with the calculator's own default shape:
- 5 missed calls a week, of which 30% were a new job: 1.5 new-job calls.
- Close 1 in 3 of them: 0.5 jobs a week.
- At $3,000 a job: $1,500 a week.
- Over 46 working weeks, the count the calculator uses after leave and stat days: $69,000 a year.
Against that, an entry AI answering plan at NZ$99 a month is $1,188 a year and the top of the band is $4,800. On those inputs the decision is not close.
Now the caveat, which matters more than the result. Every one of those four inputs is an assumption. The calculator page says so on its own face: the close rate and job value match an illustration used elsewhere on this site rather than a citable NZ statistic, and the calls-missed and share-of-new-job defaults have no source at all. Some missed calls are wrong numbers, spam, or an existing client checking on work already underway. Some callers ring back. Pull a month of your own call log, count the ones that were a new job you never heard from again, and put that number in. If the answer still clears $1,200 a year, buy the thing.
Where it breaks, and how the breakage feels
Accents first. Speech recognition has improved a long way, and it still degrades on strong regional accents, on migrant English, and on Māori and Pasifika place names and surnames. A caller from Rangiora saying a street in Ōtautahi at speed is a harder problem than a demo call from a quiet office. The usual damage is a mangled name or a wrong street rather than a dropped call, which is survivable if you get the recording alongside the message and unrecoverable if you get a summary only.
Background noise second. Your callers ring from sites. An angle grinder, wind on a roof, a ute at 100kph with the window down. Every one of those raises the error rate, and phone numbers are the worst casualty because there is no context to correct them from. A service that repeats the number back to the caller for confirmation is worth more than one with a nicer voice.
Third, the caller who wants a person. Some do, immediately, and get short with a machine that keeps asking questions. The failure mode to test for is an agent that loops instead of escalating. Ring your own number, say "can I speak to someone", and see what happens. If it takes more than one attempt to get a transfer, that is a live problem.
Fourth, an out-of-date knowledge base. The service area you changed in April, the price you no longer charge. A model states old information with complete conviction, the same failure I wrote up in why my agent kept giving confident wrong answers. Give the knowledge base an owner and a calendar reminder.
When a human answering service is the better buy
Four situations where I would spend the extra rather than automate.
After-hours emergency work. Burst pipes, gas, no power in a house with a baby in it. Those callers are stressed and the call needs judgement about severity that a script does not carry. A person who can say "I am getting someone to you" and mean it is worth the premium.
Low call volume. If you miss one call a fortnight, the missed-call sum lands under the subscription and the AI is a cost with a story attached. Divert to your mobile and get on with it.
A client base that expects a person. Long-standing commercial clients, older residential clients, work that comes by relationship. Some will hear a machine and ring the next name on the list, and you will never know. If most of your work is repeat and referral, the phone is part of the service.
Regulated or sensitive intake. Health, care, anything where what the caller says needs handling with a duty attached to it. Buy the human.
When you do not need either
Plenty of small operators answer their own phone and should keep doing it. If you are a one-person outfit taking six calls a day and you answer five of them, the missing one is a voicemail you return at smoko. There is no problem here to solve.
The same applies if your calls are almost all existing clients about jobs in progress. That is coordination rather than lead capture, and the job management software roundup is the more useful read.
And if quotes take three days to go out, answering the phone faster does nothing. Fix the quote first, which is the argument in nobody automates the quote.
Working out which one you are
Run the numbers in the missed-call calculator first, with a month of your own call log rather than the defaults. If the answer says the phone is costing you real work, book a free 30-minute call and I will go through the three vendors above against how your jobs actually arrive.