Questions I getasked every week.

It answers phone calls, chats and DMs in your business's voice, books jobs into your calendar, qualifies website visitors before they reach you, and drafts and schedules content. These are narrow, repeatable jobs with a clear right answer, which is where the technology is dependable.

Self-serve tools start around $50 to $150 a month per job. A configured system built on your prices, hours and escalation rules usually runs $300 to $1,200 a month with a setup fee. Custom builds start in the low thousands and only make sense for unusual workflows.

A self-serve tool is running the same day but needs a few weeks of correction before it is trustworthy. A configured system typically takes two to four weeks from first conversation to handling live customers, with the bulk of that time spent gathering your prices, policies and escalation rules.

Ninety days is realistic for a business with no technical staff. Roughly a month to baseline your numbers and document your answers, a month building and running the system in shadow mode where only staff see the output, and a month of supervised live operation with a widening scope.

Shadow mode means the AI processes real enquiries and produces real answers, but a person reviews and sends everything and no customer sees the raw output. It surfaces wrong prices and bad tone at zero risk, and two weeks of it removes most of the errors that would otherwise reach customers.

No software. Count your enquiries, your response times and your close rate to create a baseline, then write down your prices, service area, hours, escalation rules and the twenty questions you answer every week. That document is what any system gets built on.

A service business with two to fifty staff can cover the useful ground for under $300 a month in New Zealand dollars. Most of that goes to whatever answers enquiries, since voice and chat carry usage charges. Everything else tends to sit in the $20 to $60 range per tool.

Five jobs rather than five brands. Something to capture enquiries from every channel, something to qualify them, something to respond fast, something to help produce content, and something to measure the result. One paid tool per job is usually enough, and several tools cover two jobs at once.

On overlapping subscriptions bought at different times, on plans sized for volume they do not have, and on tools nobody was made responsible for. The second most common waste is paying for a general writing assistant when the same capability is already bundled into a tool they already have.

An AI agent is software that holds a conversation and can then take action in your other systems on the strength of it. It reads your calendar, books the slot, sends the confirmation and updates your job list, rather than only replying with words and leaving the work to you.

A chatbot produces answers. An agent produces outcomes. The chatbot tells a customer you have Thursday free; the agent checks the calendar, holds the slot, confirms by text and adds the job. The difference is access to tools and permission to use them.

They fail on emotional conversations, on anything requiring judgement about price or scope, on questions nobody wrote an answer to, and on broken internal processes. A good agent detects these cases and hands the customer to a person with the full conversation attached.

Some do by default and some never do. Business plans from major providers generally exclude customer data from training, while free and consumer tiers often do not. The setting is usually available but rarely on by default, so check it explicitly and ask the vendor to confirm the position in writing.

Ask where the data is physically stored, whether it is used for model training, how long conversations are retained and whether you can shorten that, which of their staff can read conversations, and what happens to your data if you cancel. Vague answers to any of these are a reason to look elsewhere.

It can be, provided you limit what the agent can see and do. Give it the information needed to answer enquiries rather than access to your full customer database, keep payment details and identity documents out of the conversation entirely, and make sure a human reviews anything sensitive.

Buying is cheaper for almost every small business. A subscription tool costs tens to a few hundred dollars a month with no development spend. A custom build starts in the thousands and carries ongoing maintenance. Custom only wins when a subscription tool cannot do the job at all.

Three situations justify it. Your workflow is unusual enough that no product models it. Your customer data legally or contractually cannot sit on a vendor's servers. Or you need to integrate with older software that subscription tools do not support. Outside those, buy.

Per-conversation or per-minute usage charges above the base plan, setup and configuration time, integration work to connect your calendar and CRM, and the cost of migrating out later if your data is locked in a format you cannot export.

For most service businesses it is answering inbound enquiries. The volume is high, the answers are consistent week to week, the information already exists in your head, and a mistake usually costs a follow-up call rather than a customer. Content drafting is the common second choice.

Score it on five things: how often it happens, whether the right answer is consistent, what a mistake costs, whether the information the AI needs already exists in writing, and who will check the output weekly. A task that scores poorly on two of these will not survive.

Small enough to finish in a month and visible enough that you would notice if it stopped. A project too small to matter never gets attention, and a project touching four departments never gets finished. One task, one owner, one measurable number.

Usually because the business bought a platform before defining the job it should do, or automated a process that was already broken. Technical failure is rare. The common pattern is a system that works as built but was pointed at the wrong problem, so nobody can say whether it helped.

Record your current numbers first. How many enquiries arrive, how many you answer, how long you take to respond, and how many turn into paid work. Without those figures from before launch, you cannot prove afterwards that the AI changed anything, and the subscription becomes a matter of opinion.

Yes. Every AI conversation needs a visible route to a person, triggered by an explicit request, by signs of frustration, or by a topic outside its scope. Hiding the human option is the fastest way to turn a mildly annoyed customer into one who leaves.

Containment is the share of conversations the AI finishes without a human. For a well-configured agent handling routine enquiries, somewhere between half and three quarters is realistic. A very high rate can mean the agent is refusing to escalate, so always read it alongside how those conversations ended.

Reconstruct one from what you already hold. Phone systems keep missed call logs, email keeps timestamps, and your accounting software holds job counts and values by month. A rough baseline built from records you already have beats no baseline, though a fortnight of deliberate counting is better.

Total monthly cost of the AI system, including subscription and usage charges, divided by the number of enquiries it resolved without human involvement. It gives you a single figure to compare against what the same work costs in staff time, which is the comparison that decides renewal.

Work out the revenue you currently lose to slow or missed responses, multiply by the share the system would realistically recover, then subtract the full monthly cost including your own oversight time. Divide the setup fee by that monthly gain to get a payback period in months.

Under six months is strong, six to nine months is acceptable if the underlying problem is stable, and beyond nine months the assumptions are usually doing too much work. Recalculate with half your recovery estimate before committing, because the pessimistic case is the one you should be able to live with.

When volume is low, when the process it sits on top of is broken, or when the bottleneck is downstream. A business getting eight enquiries a week, or one that already answers every call but takes nine days to send a quote, will not get its money back from an answering agent.

It answers inbound calls, website chats and social messages, identifies what the caller wants, answers common questions from your own information, books or reschedules appointments in your calendar, and passes anything outside its rules to a person with the full conversation attached.

Many can, and the good systems say so at the start rather than pretending otherwise. Callers rarely mind an AI that answers immediately and knows the answer. They mind an AI that stalls, loops or refuses to connect them to a human.

A basic setup answering questions and taking messages can run within a week. Adding calendar booking, customer records and escalation rules usually pushes it to two or four weeks, most of which is you deciding the rules rather than anyone writing code.

Yes. It answers on your existing number, asks what the problem is, checks whether you cover the suburb, gives your standard callout price, and either books a slot or flags the job as urgent for you to ring back. Anything unusual gets passed straight to you.

It can give the price bands you set, such as a callout fee or a typical range for a common job, and it should say clearly that anything else needs a site visit. Never let an agent quote a variable job, because you will be held to whatever number it said.

You set the rules. Most trades have the agent ask two or three triage questions, then ring or text your mobile immediately for anything meeting the emergency definition, while non-urgent jobs get booked into the next available slot without waking you.

No, and it must be configured so it will not try. An AI agent handles bookings, reminders, practice information and fee ranges. Any question about symptoms, pain, medication or treatment suitability has to be routed to a clinician, with the agent saying plainly that it cannot advise.

It confirms appointments through the channel the patient actually reads, usually text, offers a one-tap reschedule instead of a silent cancellation, and immediately offers the released slot to patients on a short-notice list. The reschedule path is what converts a would-be no-show into revenue.

Yes, provided you meet your obligations under the Privacy Act and the Health Information Privacy Code. That means collecting only what you need, telling patients how their information is used, storing it securely, and knowing where your vendor holds and processes the data. Seek your own advice on your specific setup.

Yes, by a wide margin on the monthly cost. A part-time receptionist in New Zealand costs several thousand dollars a month once you include holiday pay and KiwiSaver, while a configured AI system typically runs in the hundreds. The comparison changes if you need judgment rather than coverage.

Read tone and respond to distress, make a commercial judgment call about squeezing in a job or discounting, handle a complaint without escalating it further, and recognise a regular customer's context without being told. These are the conversations worth protecting from automation.

An answering service gives you a real person but usually only takes a message, because they do not know your prices or availability. An AI receptionist answers the question and books the job but is weaker on emotionally charged calls. Volume and complexity should decide it.

Automated replies are allowed through Meta's official business messaging APIs, within rules that govern consent and when you may message someone. Unofficial tools that log into your account to send messages breach the terms and put the account at risk. Check the current Meta business messaging policy before you configure anything.

When a customer messages your business, Meta opens a period during which you can reply freely in conversation. Once it closes, further outbound messages are restricted to approved template types, and on WhatsApp these are usually paid. The exact duration and categories change, so confirm them in Meta's current documentation.

Yes. A DM asking your price is the same lead as a phone call asking your price. Routing every channel into one inbox and one CRM stops duplicate follow-up, gives you a true count of enquiries, and lets the same knowledge base answer consistently everywhere.

A current service list with prices or price bands, your service area and hours, written answers to the questions customers ask most often, clear escalation rules for what it must not handle, and a short note on tone. That set covers the overwhelming majority of real conversations.

You can, and the answers usually get worse. Marketing pages are written to persuade rather than to answer, and old pages contradict new ones. A short curated knowledge base written as direct answers outperforms a full site crawl almost every time.

Gathering the source material takes most owners two to four hours spread over a week. The first configuration takes a day or two. The part that matters is the following fortnight of reading transcripts and correcting answers, which is where accuracy actually comes from.

It holds the slot the moment it offers it, then writes the confirmed booking back to the same calendar the rest of your team uses. Double bookings happen when the agent reads a stale copy of the calendar or when staff book in a second system it cannot see.

Yes, if you give it the rule. It needs job durations, a travel buffer between locations, and ideally a distance check between the previous booking and the new address. Without those inputs it will happily book two jobs an hour apart in suburbs forty minutes apart.

Anything where a wrong duration is expensive or unsafe. Large or unusual jobs, first appointments needing a site visit to scope, anything with a clinical judgement attached, and any booking that would move or displace an existing customer. Those go to a human.

It varies by industry, but service businesses commonly see a lift in the early evening and across weekends, because people search for a plumber or a physio once they have finished their own working day. The only reliable answer is the timestamp data from your own phone system and web forms.

Rarely. Someone with an urgent problem tends to keep dialling the next business on the list rather than leave a message. Voicemail catches the patient callers and loses the ones in a hurry, which are often the higher-value jobs.

Yes for the common cases: answering questions about services, pricing bands and availability, collecting contact details and booking into an open slot. It needs an escalation path for anything urgent or unusual, and a clear rule about when to wake a human.

On three triggers. When the customer asks for a person, immediately and without negotiation. When the conversation shows signs of frustration, such as repeated rephrasing or short sharp replies. And when the topic falls outside what the agent has been given permission to handle, such as complaints, refunds or anything with legal or clinical weight.

The full conversation transcript, a one-line summary of what the customer wants, any details already collected such as name, address and job type, and the reason for the escalation. The human's first message should show they have read it rather than asking the customer to start again.

Say so honestly and commit to a specific time. A message like 'nobody is free right now, I'll have someone call you before 10am tomorrow' is far better than silence or an open-ended promise, provided the callback actually happens. Then log it so the follow-up is not left to memory.

Expect roughly $50 to $250 a month for a self-serve tool you configure yourself, $400 to $1,500 a month plus a setup fee for a system built around your business, and $8,000 upward as a project for a custom build. Voice adds per-minute charges on top.

Voice bills for speech recognition, the language model and speech synthesis for every minute of every call, and it needs telephony to carry the call. Chat bills per conversation or per message and has none of that audio processing, so the same volume costs considerably less.

The usual surprises are setup and onboarding fees, per-integration charges to connect your calendar or CRM, overage rates once you pass an included usage tier, and change fees for editing the script after launch. Ask for all four in writing before signing.

Check your phone system's call reporting if you have a VoIP provider, or scroll your mobile call log and count red entries. Count calls that rang out while you were on another call as missed too, because the caller experienced them the same way.

Multiply your monthly missed calls by the share that are genuine enquiries rather than sales calls, then by the rate at which you win enquiries, then by your average job value. That gives you monthly lost revenue. Multiply by twelve for the annual figure.

Some do, particularly existing customers and anyone who specifically wants you. New enquiries with an urgent problem generally work down the list of search results until someone answers, so the callback rate on first-time enquiries is much lower than owners expect.

Because most of them are obstacles rather than help. The common complaints are being trapped in a loop with no way to reach a person, receiving an answer to a question nobody asked, and being made to repeat everything once a human finally appears. A chatbot that resolves the question gets no complaints.

No. Customers usually work it out within two messages, and the discovery costs you more trust than the disclosure ever would. State plainly that it is an assistant, offer a route to a person, and let the quality of the answers do the rest.

The option should be visible from the first message and honoured the instant it is asked for. As a rule, if the bot has failed to help twice on the same question, it should offer a person without being asked rather than trying a third time.

It is a website with a layer on top of the pages that responds to visitors in real time. It answers questions using your own prices and policies, asks qualifying questions, and books a time in your calendar, instead of only displaying information and waiting for a form submission.

Usually not. Most of the capability sits on top of your existing pages as a chat layer, a booking integration and better calls to action. A rebuild is only worth it when the site loads slowly, cannot be edited, or the page structure itself is confusing visitors.

No. Keep the form for people who prefer it, and for anyone who arrives after a long research session already knowing what they want. The AI layer catches the larger group who have a question first and would otherwise leave without asking it.

Choose a tool, load it with your services, prices, service area and escalation rules, then install its script tag on every page. Most platforms give you a single line of code to paste before the closing body tag, or a plugin if you run WordPress. Budget a day for the content, not the install.

AI chat wins on coverage and response speed, since it answers instantly at any hour. Live chat wins on complex or high-value conversations. Most small businesses end up running AI chat with a clear route to a human, which covers the volume without needing someone watching a queue.

Only if it is configured badly. The common mistakes are popping open within two seconds of arrival, covering the content on mobile, sending follow-up chimes, and hiding the option to reach a person. Delay the invitation, keep it small on phones, and make the human route obvious.

It can, and the usual cause is changed URLs with no redirects rather than the design itself. If every old page either keeps its address or has a 301 redirect to the closest equivalent, and the page content survives the move, rankings normally hold through a rebuild.

Record monthly organic traffic, your most visited pages, total enquiries by channel, and load time on a mobile connection. Without these, you cannot tell afterwards whether the new site helped, and most businesses discover this only when it is too late to check.

For a service business with 10 to 30 pages, expect six to twelve weeks from brief to launch. Content is almost always what causes delays, not build time, so having copy and photos ready before design starts is the single biggest schedule saver.

It is a system that talks to a visitor before you do, asking a few useful questions about what they need and when, then scoring and labelling the enquiry. You still receive every lead, but each one arrives with context so you know which to ring back first.

Only if they read like a form. Questions that give the visitor something back, such as a price band or an availability window, tend to increase completion rather than reduce it. Asking for a budget figure with no explanation is what drives people away.

Three or four is usually enough to sort your enquiries reliably. Aim to capture what the job is and roughly when the person needs it done. Everything beyond that can wait until you speak to them directly, and each extra question costs you completions.

A complete and active Google Business Profile, a separate page for each service you sell, a steady flow of genuine reviews, and content that answers the questions customers actually ask. Technical work matters less than these once your site loads reasonably and can be crawled.

A page per service is worth building, because each one answers a different search and a different customer question. Pages per suburb only work if each has genuinely different content. Twenty near-identical location pages tend to underperform a handful of strong ones.

Reviews are part of how Google assesses local businesses, and they also influence whether someone clicks your listing over a competitor's. Ask every satisfied customer, respond to what comes in, and never buy or incentivise reviews, which breaches Google's policies.

Slower pages lose visitors before they see anything, and the loss compounds because every later step depends on the visitor still being there. The relationship is directional rather than a fixed formula, so measure it on your own site by comparing conversion rates between fast and slow landing pages in your analytics.

They are Google's three page experience measures: how quickly the main content appears, how fast the page responds to a tap, and how much the layout shifts while loading. They matter because they describe what a real visitor experiences on a phone, and Google uses them as a ranking signal.

Run your key pages through Google's PageSpeed Insights and read the mobile results rather than the desktop ones. Then open the site on your own phone, on mobile data with wifi turned off, in a private browsing window so nothing is cached. That second test is the one that reflects your customers.

Website personalisation means changing what a page shows based on something you know about the visitor, such as which page they arrived on, what time it is, or whether they have been to the site before. The page content adapts while the underlying offer stays the same.

No. The most useful signals are first-party ones that come from the current visit: the referring link, the page path, the time of day, and whether a cookie from a previous visit exists. None of that requires third-party tracking or buying data about individuals.

It is worth it once your pages are already converting reasonably and you have enough traffic to tell whether a change helped. Before that point, fixing load speed and page clarity will move more enquiries than any adaptive content will.

For a small service business, expect to pay between $2,000 and $8,000 for a professionally built site of five to fifteen pages. Below that you are usually buying a template with your logo on it. Above $15,000 you should be getting custom functionality rather than just design. Prices vary widely by agency and scope.

Budget roughly $300 to $1,500 a year for hosting, domain renewal and security updates on a standard small business site. Add more if you want regular content changes, and add a monthly fee if you run an AI chat or booking layer. Ask for these figures in writing before you sign anything.

Most of the variation comes from how much of the work is custom, how much content the agency writes for you, and whether integrations are included. A template site with client-supplied copy and a bespoke build with written content and a booking integration are different products, even though both are described as a website.

Long enough to answer every question a buyer would ask before ringing you, which for most service businesses is between 600 and 1,200 words. Length itself does not help or hurt. Pages fail when they are padded with adjectives rather than when they are long.

Put a range or a starting figure on the page, even if every job is different. Visitors who cannot find any price signal assume the worst and leave. A stated range filters out the people who were never going to buy and reassures the ones who were.

It should name the service and the place, then add the thing the buyer most wants to know. Something like 'Emergency plumbing in Auckland, someone on the way within two hours' works because it confirms the visitor is in the right place and answers their real question at once.

For a local service business, somewhere between two and five percent of visitors making contact is a reasonable working range. Below one percent usually points at a specific problem rather than bad luck. Measure your own baseline first, because industry averages hide enormous variation by traffic source.

The most common causes are a slow page on mobile, an unclear offer, no price signal anywhere on the site, and a contact option buried below the fold. Traffic quality matters too. Visitors arriving from unrelated searches were never going to enquire regardless of the page.

Set up conversion tracking in your analytics for form submissions, phone taps and chat starts, then compare enquiries against sessions for the last ninety days. Without that baseline you cannot tell whether a change helped, and most sites have never had it configured.

No. AI can plan a month of posts, draft captions in your voice and answer routine comments within seconds. It cannot photograph your work, judge whether a post is on brand, or handle a complaint. Expect it to remove roughly half the work rather than all of it.

Platforms rank posts on how people respond to them, not on which software typed the caption. Reach falls when the content is generic, repetitive or slow to reply to. AI-assisted posts that carry real photos and specific detail perform like any other post.

A do-it-yourself stack of a scheduling tool and an AI writing subscription usually runs under $150 a month. A managed setup where someone builds the voice document, the calendar and the reply rules for you sits higher, and should be judged against the hours it gives back.

Two or three times a week is enough for a local service business, and consistency matters far more than volume. An account posting twice weekly for a year outperforms one that posts daily for six weeks and then goes silent, because most people check whether you are still active before they enquire.

No. A recent phone in daylight is fine, and for trades and clinics an unpolished real photo often outperforms a styled one. Wipe the lens, stand so the light is behind you, and take the before shot before you start work rather than trying to recreate it later.

It rarely generates cold enquiries on its own. It converts people who already heard your name, by showing recent work and proving you are still trading. Treat it as the reference check that happens before someone rings, and measure it by asking callers where they found you.

Early engagement on a post signals to the platform that the content is worth showing to more people, and a reply from the account restarts that engagement. A comment answered within the first hour tends to attract further replies. Answering the same comment two days later does almost nothing for distribution.

They notice when the reply ignores what they asked, uses language you never use, or refuses to hand over to a person. They rarely notice a short, accurate answer to a common question. Keep automated replies factual and brief, and escalate anything with emotion or money in it.

Complaints, anything about a specific job already underway, health or safety questions, pricing on non-standard work, and anything from a person who is upset. Those need a human who can take responsibility. Automate the questions where the answer is the same every time regardless of who is asking.

One month is the practical horizon. Anything longer goes stale before it publishes, and anything shorter means you are writing under pressure. Plan thirty days in a single sitting, then spend fifteen minutes each week approving the days ahead and swapping in anything topical.

Three posts a week is enough for most local service businesses, and two done consistently beats five done for a fortnight then abandoned. Frequency matters far less than being findable and responsive when someone checks whether you are still trading.

For each slot record the date, the pillar it belongs to, the angle in one line, the photo or clip it needs, and the caption. The photo requirement is the field most calendars leave out, and it is the one that causes posts to be missed.

Google has said it rewards helpful, original content regardless of how it was produced, and demotes content made mainly to game rankings. In practice thin, repetitive pages lose visibility whether a person or a model wrote them, and useful pages hold it either way.

Detection tools produce both false positives and false negatives often enough that no serious decision should rest on their score. Plain, formulaic human writing frequently registers as AI-generated. Human readers judging by feel are more reliable than the software claiming to measure it.

Using AI to draft is fine. Publishing unedited output is the problem, because the default style is generic enough that readers register it as filler. Add a specific detail only your business knows and cut the formulaic phrasing, and nobody minds how the first draft appeared.

Ask every caller how they heard about you and write the answer down in the same place every time. Pair that with a dedicated landing page used only in your social profiles, so clicks are counted separately. Neither is perfect alone, but together they give you a defensible picture.

Expect three to six months of consistent posting before enquiry patterns become readable. Social rarely produces same-day sales for service work. It builds familiarity that shortens the decision when someone finally needs you, which shows up as a slow lift in direct and branded enquiries.

Only as an early warning signal. Follower count does not correlate with revenue for a local service business serving one city. Track saves, shares, profile visits and link clicks instead, because those show intent to act rather than passive scrolling.

No. A recent phone shoots better video than most people can light or frame. Spend your money on a small clip-on microphone instead, because bad audio loses viewers faster than soft footage does. Film in landscape only if you plan to crop later.

Between 15 and 40 seconds for most service businesses. That is long enough to show a before state, the work, and the result, and short enough that people finish it. Watch-through rate matters more than length, so cut anything that does not move the clip forward.

AI handles the mechanical parts well: transcribing, generating captions, removing silences, reframing to vertical, and suggesting clip boundaries from a longer recording. It handles taste badly. Pick your own hook and your own opening frame, then let the tool do the tedious work.

Taking one piece of thinking and reshaping it for different platforms and formats, so a single hour of work produces a week or more of output. Done properly it changes the angle and the framing for each destination rather than copying the same text into several apps.

Your inbox and your phone. Any question a customer asked more than twice is a proven idea, because someone already cared enough to ask it. Questions beat trends for local service businesses, since they map directly onto what people search for when they need you.

Cross-posting identical text rarely gets penalised outright, but it performs poorly because each platform's audience expects a different register. A LinkedIn audience and a TikTok audience want different framing of the same fact, and the version written for neither tends to land with neither.

Write a one-page voice document containing the words you use, the words you never use, your typical sentence length, how formal you are with strangers, and three samples of your own writing. Paste it at the top of every request. Samples do more work than instructions.

It produces the most average phrasing available, because that is the safest output across every possible reader. Averageness reads as polish in isolation and as emptiness across a whole feed. Specific instructions and real writing samples pull it away from the average.

One page. Anything longer gets partially ignored and is harder for you to keep current. The useful content is a banned word list, two or three rules about sentence construction, and three samples of writing you were happy with.

Most Auckland service businesses spend between $200 and $900 a month on a working AI setup, plus a one-off configuration cost of roughly $1,500 to $6,000 depending on how many systems it connects to. Voice answering sits at the higher end because calls are billed per minute.

It depends on enquiry volume rather than headcount. If you take more than about twenty enquiries a week and miss a meaningful share of them, the arithmetic usually works. If you take five enquiries a week and answer all of them, it will not.

Modern speech models handle New Zealand accents well. Local place names are the weak spot, so suburb and street names need to be added to the agent's vocabulary during setup and tested with real recordings before it goes live.

It can be, with the right vendor settings. Check that your data is excluded from model training, ask where it is stored and for how long, and confirm who at the vendor can read it. Put those answers in writing before any client information goes near the tool.

Document chasing. Reminding clients to send bank statements, receipts and signed forms is high volume, low judgement and entirely rules-based. It is also the task most likely to be delaying your team's actual work during a busy filing period.

No, and you should not let it. AI can draft client emails, answer routine questions about deadlines and process, and assemble information for review. The preparation, the professional judgement and the filing remain the responsibility of the qualified person signing off.

It answers the phone during service, takes and changes bookings, replies to enquiries about hours, parking, dietary options and group sizes, sends confirmation and reminder messages to cut no-shows, and drafts replies to online reviews for an owner to approve before posting.

A booking platform with automated confirmations typically runs from around $50 to $200 a month depending on covers. An AI phone agent that answers calls and takes bookings usually sits in the low hundreds per month. Setup is often a one-off fee on top.

Only if it hides the option to reach a person. A well set up agent answers on the first ring, handles the simple booking or question in under a minute, and transfers anything unusual to the venue. Regulars generally prefer that to a phone nobody picks up.

No. An AI agent in a clinic should handle bookings, reminders, hours, pricing and general practice information only. Any question about symptoms, treatment or whether someone should be seen must go to a qualified practitioner. Build that refusal into the system before it goes live.

No-shows and late cancellations, mostly. Automated reminders with easy rescheduling recover appointments that would otherwise be lost, and an agent answering after-hours enquiries books people who would have rung a competitor the next morning.

Health information collected by a health agency is covered by the Code, and using an AI tool does not change that. You still need to justify what you collect, keep it secure, limit how long you hold it and understand where an overseas provider stores it. Take specific advice for your practice.

It replies to trial and membership enquiries within seconds at any hour, books the first session, runs the onboarding message sequence for new members, and flags members whose attendance has dropped so a coach can reach out before they cancel.

It can surface the signal early. Most cancellations follow weeks of declining attendance, and that drop is visible in your check-in data. Automation spots the pattern and prompts a human to make contact. The contact itself still needs to come from a person the member knows.

Most studios spend between $100 and $400 a month across a membership platform with automated messaging and an AI agent handling enquiries. Weigh that against the value of a single retained member, which for most studios is several hundred dollars a year.

Use is uneven. Most owners have tried a writing assistant informally, far fewer have anything connected to their phone, website or booking system. The gap between casual use and a system that handles customer enquiries is where almost every New Zealand business currently sits.

It depends on your registration status and the supplier. Many overseas providers stop charging GST once you supply a valid New Zealand GST number, while others charge it regardless. Check each invoice and confirm the treatment with your accountant rather than assuming.

There is no blanket requirement for personal information to be stored in New Zealand. The Privacy Act does place obligations on you when you disclose personal information to an overseas party, so you need to know where a vendor stores data and what protections apply.

Responding to portal and website enquiries within seconds instead of hours. Buyers and vendors enquire on several listings at once, and the agent who replies first usually gets the conversation. Everything else AI does for an agent is worth less than that.

It can produce a solid first draft from your property details, which saves most of the writing time. You must check every factual claim yourself, because advertising accuracy obligations sit with you and an AI draft will confidently state things about a property that are not true.

Not by itself. You do need to know where the vendor stores contact details, how long they are kept, and whether conversations are used to train the provider's models. Be upfront with people that an automated assistant handles first contact.

It answers the calls and messages you cannot get to, works out whether the job is urgent, collects the address and the problem, and books it into your calendar. It then follows up quotes that went quiet. The rest of the pitch you will hear is mostly noise.

A configured setup that answers calls and books jobs typically costs $1,500 to $5,000 to set up and $250 to $700 a month to run in NZD, with voice at the higher end because phone minutes are billed per minute. Chat-only setups cost considerably less.

Most will work it out, and that is fine as long as you are upfront and the agent is fast and accurate. What annoys people is a system that pretends to be human, loops them in circles, or refuses to put them through to a person when they ask.

More than the hourly rate. On top of gross wages you carry the compulsory employer KiwiSaver contribution, four weeks of annual leave, paid public holidays, sick leave, ACC levies, payroll administration and equipment. Recruitment time and a training ramp of several weeks sit on top of that in year one.

When the work is high volume, well documented and needs to happen outside business hours. Answering the same enquiries at 9pm on a Sunday is a clear automation case. Work that requires judgement, relationships or physical presence is not.

When the process is not documented, when volume is low and irregular, when the role needs someone to own outcomes and improve the process, or when the work involves judgement calls and difficult conversations. A person also absorbs new tasks a system cannot.

Yes. New Zealand's Privacy Act applies to every agency handling personal information, including sole traders and small businesses. There is no exemption based on turnover or staff numbers, which is a common assumption people carry over from other countries.

Generally yes, but sending personal information to an overseas provider is treated as a disclosure and carries obligations around the protections that apply at the other end. Check where the vendor stores data and what its contract says before you start, and seek advice if the information is sensitive.

Being upfront is the safer position and it is what the Office of the Privacy Commissioner encourages in its guidance on automated tools. Beyond privacy, telling people plainly avoids the trust damage that comes when a customer works it out for themselves partway through a conversation.