Conversational AI, minus the enterprise brochure.
Every conversational AI website is written for a bank. Omnichannel this, digital transformation that, a Gartner quadrant somewhere near the bottom.
This one is written for a bloke with four utes and a phone that rings while he’s under a sink. What the technology actually is, what the types of it are, who sells it, and which parts are any use to an Australian trade business.
Ring the number. It’s a real conversational AI, answering as a real business.
What conversational AI actually is.
Conversational AI is software that can hold a conversation in ordinary language — spoken or typed — work out what the person actually wants, and then go and do something about it. That last part is the whole difference. Plenty of things can talk. Not many can book a job.
Conversational artificial intelligence is the formal name for it. Underneath, the conversational AI technology is four separate jobs stitched together, and it helps to know which one is failing when something goes wrong:
Hearing you
Speech recognition turns what the caller said into text — through a bad line, a running tap, a dog, and an accent the software has never met.
- Where most cheap AI talking bots fall over first
- Australian suburb names are their own special problem
Understanding you
Natural language understanding decides what the words mean. “My hot water’s gone”, “no hot water” and “the shower’s freezing” are one intent, not three.
- This is the bit that used to need a scripted decision tree
- Large language models are why it finally works
Doing something
Checking the diary, quoting the right after-hours rate, creating the job. A chat bot that only answers questions is a brochure that types.
- Needs a real connection into your job system
- The step most free AI chat tools simply don’t have
An old chat bot follows a script somebody wrote. Conversational AI works out what you meant, even when you said it a way nobody anticipated — and then acts on it.
Everything else on this page is detail hanging off that difference.
Same technology, three very different products.
Conversational AI for business splits into three things that share a brain and almost nothing else. All three get called an AI chatbot at some point, which is why comparing vendors is so confusing. They’re not really competing — they’re answering different questions.
Voice — the phone
A voicebot or AI receptionist that answers the phone, talks like a person, and books the job. For a trade business this is the one that pays, because the phone is where your money arrives.
- Hardest to build — latency and interruptions are brutal
- What we build
AI website chat
The AI chat bot for a website — the bubble bottom-right that replaced live chat. Good at answering questions at 11pm, capturing a name, and pushing people towards a form.
- Cheap and quick to put on an existing site
- Only works on the customers who visited the site
Messaging & SMS
The same brain over SMS, WhatsApp or Messenger. Handy for confirmations, running late texts, and getting a photo of the fault before you drive out.
- Best as a follow-up channel, not a front door
- Where conversational commerce actually lives
Vendors sell “one conversational AI across every channel” as the goal. For a business with a handful of vans, it’s a distraction. Ninety per cent of the money walks in through the phone. Fix that first, and add the website chat later if you still want it.
Six things that all get called “AI chat”, side by side.
A conversational AI chatbot on a website and an AI voice receptionist on the phone run on much the same engine. What separates them is what happens next — and that’s easiest to see against one call. The test case is the one your business actually gets: a caller at 7pm saying “there’s water coming out from under the hot water thing.” Here’s what each option does with that.
| What it is | How it works | What it does with the 7pm call | Who it suits | Books a job? |
|---|---|---|---|---|
| Rule-based chat botThe old kind. Buttons and a decision tree. | A script someone wrote. Every path anticipated in advance. | Nothing — it isn’t on the phone. On the website it offers four buttons, none of which say “water everywhere”. | Simple FAQ deflection on a website with steady, predictable questions. | |
| Free AI chatChatGPT, Gemini, Copilot and the rest. | A general-purpose large language model in a chat window you go to. | Gives the caller good general advice about isolating a hot water system. Doesn’t know you exist, doesn’t know your rates, doesn’t ring. | You, personally — writing quotes, drafting emails, working things out. | |
| AI chat bot for your websiteThe DIY widget you install yourself. | An LLM pointed at your website’s content, in a bubble on your site. | Answers well if the customer happens to be on your website at 7pm with water rising. Most aren’t. They’re on the phone. | Businesses whose enquiries genuinely start online — retail, ecommerce, courses. | |
| Enterprise conversational AI platformCognigy, Kore.ai, Yellow.ai, Ada, Amelia, Genesys and similar. | A build-it-yourself platform with flow designers, analytics, and a contact centre bolted on. Priced per seat or per session. | Anything you want — after your team, or an implementation partner, spends months building and testing it. | Banks, telcos, airlines, health funds. Hundreds of agents, a project team, a budget. | |
| Open source conversational AIRasa and friends, on GitHub, mostly Python. | Free code, your servers, your problem. You supply the speech, the model, the hosting and the maintenance. | Whatever you built it to do, once you’ve built it. Then again every time something upstream changes. | Developers, and companies with real engineering capacity and a reason to own the stack. | |
| Johnni AIVoice-first conversational AI, built for Australian trades. | We build it for your business — your rates, your service area, your rules — and maintain it. You don’t touch a flow designer. | Answers on the second ring, talks them through the isolation tap, quotes your real after-hours rate, and books it into your job system with the notes attached. | Trade and service businesses in Australia. Typically 1–30 vans. |
Categories, not a vendor scorecard — the platforms named are examples of a category, and their capabilities are broadly what their own public sites describe. If your business really is a contact centre, the enterprise row is the right one to go read about, and we’d tell you so on the phone.
Why every list of the best conversational AI companies is useless to you.
Search for conversational AI companies and you get the same names — the enterprise platform vendors, the contact-centre suites, and the open source frameworks underneath them:
They’re enterprise conversational AI platforms. They’re sold to organisations with a contact centre, an IT department, a procurement process and an implementation partner. The pricing is per seat or per session with an annual commitment, and the product you buy is a toolkit — the building is still ahead of you.
That’s not a criticism. It’s just a different job. If you run six vans out of a shed in Toowoomba, buying a platform means becoming the person who maintains a conversational AI, on top of everything else you already do.
And about the Gartner Magic Quadrant.
Gartner publishes a Magic Quadrant for enterprise conversational AI platforms, and Forrester runs its own equivalent. Both are genuinely useful documents — for the buyer they’re written for.
They rank vendors on things like breadth of channel support, developer tooling, analytics depth and global partner networks. None of those are what decides whether the phone gets answered at your business at 7pm on a Sunday. Nobody has ever lost a hot water job because their vendor’s partner ecosystem was thin.
If you’re a trade business, the honest shortlist criteria are much shorter: does it answer, does it sound like a person, does it know your prices, and does the job land in the system you already use.
Free AI chat is brilliant. It just isn’t a receptionist.
ChatGPT, Gemini, Claude and Copilot are free or near enough, and you should use them. Quotes, emails, ad copy, working out what a manufacturer’s spec sheet actually means — genuinely useful, today.
But a free AI chat window is somewhere you go. A receptionist is something that comes to you. Anyone can chat with an AI online for nothing, from any phone, right now — and the gap between that and “my phone gets answered” is four things, none of which are the clever bit:
A chat window can’t pick up a phone. Somebody still has to, or it goes to voicemail — and most people don’t leave one, they ring the next business.
Your after-hours rate, the suburbs you’ll drive to, the jobs you won’t take, the brands you won’t touch. A general chat model guesses. Guessing about price is expensive.
Offering a time means reading real jobs from your scheduling system, not inventing a plausible-sounding slot.
The conversation ends and the information evaporates. No job card, no customer record, nothing for the morning.
The model is the commodity. The phone line, your business rules, the diary and the job system are the work.
You can build one yourself. Here’s what it really costs.
The parts are all public. Open source conversational AI frameworks like Rasa, models on Hugging Face, speech-to-text and text-to-speech APIs, a telephony provider, and a few hundred lines of Python glue. There are working tutorials on GitHub. None of that is a trick.
What people underestimate isn’t the build. It’s the second month. A voice conversation is unforgiving in ways a chat widget never is — half a second of extra latency and the caller talks over the top of it, or worse, hangs up thinking the line’s dead.
Then a model version gets deprecated, a phone provider changes an endpoint, and somebody rings on a bad line from a roof cavity. Nobody notices, because a broken AI receptionist doesn’t send an alert. It just answers slightly worse for a fortnight.
If you’ve got a developer and the appetite, build it. If your day already ends at 7pm covered in someone else’s drain water, buy it.
Wire an LLM to a speech API and a phone number. It’ll talk. It’ll be impressive.
Barge-in, silence detection, people who say “um” for three seconds mid-address.
Reading real availability and writing a job back, without creating a fourth copy of the same customer.
Every model change, every API deprecation, every weird call. At 7pm. On a Sunday.
In a trade business, it lives on the phone.
Every industry gets its own pitch. There are whole websites about conversational AI in banking, conversational AI in healthcare, conversational AI in retail, conversational AI in insurance, conversational AI in HR and conversational AI for real estate. All real, all solving a problem that isn’t yours.
The trade version is specific, and it’s this: the phone rings while both hands are full, and the caller who doesn’t get an answer rings the next name on the list instead.
Every call, second ring
Including the ones at 6am, during smoko, on a Sunday, and the three that come at once while you’re in a ceiling.
Asks what you’d ask
Is it isolated, is it one fixture or the whole house, is it clean water or waste. The answers come through on the job, not in a callback.
Your prices, not a guess
Callout fees, after-hours rates, and the work you don’t take — a licensed job you won’t touch gets said plainly rather than fudged.
Into the system you use
Times offered from real availability, and the job written back with the triage notes attached.
If your enquiries genuinely arrive as website chats rather than phone calls, a website chat bot will serve you better and cost less. If you’re a contact centre with fifty agents, go and read about the enterprise platforms above. And if your call volume is two a week, the maths doesn’t work yet — ring and we’ll say so.
If you’d rather skip the theory: the AI receptionist for trades, the plumbing version, or an honest comparison of what’s selling in Australia.
The words vendors use, in English.
Enough to read any conversational AI website without having to nod along.
The thing that reads what was said and works out a sensible reply. GPT, Gemini and Claude are the well-known ones. It’s why modern AI chat feels like a conversation instead of a form.
Natural language understanding: deciding what the person actually wants. The “intent” is the want itself — book a job, get a price, complain — regardless of the words used.
Turning audio into text. On a phone call, with background noise, this is harder than the clever language part and fails more often.
Turning the reply back into a voice. Good TTS is the difference between an AI that sounds human and one that sounds like a train station.
Whether you can talk over the top of it and have it stop and listen. People interrupt constantly. A bot that doesn’t allow it feels like a phone menu.
The pause before it answers. Over about a second, callers assume the line dropped. It’s the single biggest tell of a cheaply built voice AI.
Feeding the model your real information — prices, service area, policies — so answers come from your business instead of the model’s general knowledge.
When a model states something false with total confidence. In a receptionist, this is a quoted price nobody agreed to. It’s controlled by hard rules, not hope.
The rules it can’t cross — won’t quote outside a range, won’t promise a same-day slot, won’t take a job type you don’t do. Boring, and the reason it’s safe to leave alone.
One assistant across phone, web chat, SMS and social. An enterprise goal. For most small businesses, one channel done properly beats four done badly.
Analysing recorded conversations for what customers keep asking, where calls go wrong, and which enquiries turn into jobs. Sold separately by most vendors.
Completing a sale inside the conversation rather than sending someone to a website. Big in retail. In trades, the equivalent is the booking.
Handling a caller in another language, or switching mid-sentence. Worth having in parts of Australia, worth nothing in others.
Intelligent virtual assistant, virtual agent, smart chat bot, digital worker — vendor names for the same idea. If the underlying thing understands language rather than following a script, it’s conversational AI whatever it’s called.
A decision tree with buttons. Predictable, cheap, and instantly annoying the moment a customer wants something the author didn’t anticipate.
Gartner’s grid of enterprise vendors, and Forrester’s Wave is the equivalent. Useful if you’re buying an enterprise platform. Irrelevant if you’re buying an answered phone.
Straight answers
What’s the difference between a chatbot and conversational AI?
A traditional chat bot follows a script. Somebody wrote out the paths in advance, and if the customer steps off one of them, it breaks or repeats itself.
Conversational AI works out the meaning of what was said, even when it’s phrased in a way nobody anticipated, and can carry context across the conversation — so when someone says “the other one’s doing it too” three sentences later, it knows what “the other one” is.
Is conversational AI the same thing as ChatGPT?
No, though ChatGPT is one form of it. OpenAI’s ChatGPT is a general assistant in a chat window that you visit. Conversational AI as a business product is that kind of language understanding wired into a channel your customers already use — usually the phone — with access to your prices, your diary and your job system.
The language model is roughly the commodity part. The plumbing around it is the product.
Can I just use a free AI chat bot on my website?
You can, and for some businesses it’s a sensible first move. There are free and cheap widgets that will read your website and answer questions on it reasonably well.
The catch for a trade business is that it only helps the customers who are on your website. Most of yours are on the phone with a problem in front of them, and a website chat bot has no way to help those people at all.
What does conversational AI cost in Australia?
It splits by category. Website chat widgets run from free to a few hundred dollars a month. Enterprise conversational AI platforms are quoted per seat or per session with an annual commitment, and the implementation usually costs more than the licence.
A done-for-you AI voice receptionist sits in between, priced on how much you use it. Ours is set out on the cost page, including what it costs when a month is quiet.
Which conversational AI company should a small Australian business pick?
Not the ones on the analyst lists — those are built for organisations with a contact centre and a project team. For a business under about thirty staff, the shortlist is whoever will build it for you, keep it running, and connect it to the software you already use.
We’ve written an honest comparison of the AI receptionists selling in Australia, including where we’re the wrong choice. We’re on that list, so read it with that in mind.
Does the Gartner Magic Quadrant for conversational AI matter to me?
Almost certainly not. It ranks enterprise platform vendors on criteria that matter to enterprise buyers — channel breadth, developer tooling, analytics, partner ecosystems.
None of that predicts whether your phone gets answered properly at 7pm. If you’re spending under six figures a year, the quadrant is interesting reading and nothing more.
Is there an open source conversational AI I could build on?
Yes — Rasa is the best known, there are plenty of projects on GitHub, and the models on Hugging Face are free to use. Most of it is Python, and the tutorials work.
The build isn’t the expensive part. Owning it is: voice latency tuning, interruption handling, phone provider quirks, and every model deprecation from then on. Worth it if you have a developer, painful if you don’t.
Will callers know they’re talking to an AI?
Some do, some don’t, and we don’t think you should try to hide it. Ours introduces itself as the business’s assistant. What people actually object to isn’t AI — it’s being stuck in a phone menu or dumped to voicemail.
Ring 07 4519 7035 and judge for yourself, then decide whether your customers would mind.
What happens when it doesn’t understand someone?
It says so and transfers, or takes a proper message with a callback number — rather than guessing and booking a job that doesn’t exist.
Anyone selling you a conversational AI that never fails is selling you something else. The honest question is what it does on the way down, and whether you find out about it.
Can conversational AI actually book the job, or just take a message?
Both exist, and the difference is most of the value. Taking a message means someone still rings back tomorrow, by which time a fair few callers have booked someone else.
Booking means reading real availability out of your scheduling system, offering times that fit the run you’ve already got, and writing the job back with the triage answers attached.
Does it work for industries other than trades?
The technology does. Conversational AI in banking, conversational AI in healthcare, conversational AI in retail and conversational AI in insurance are all established, generally delivered through the big enterprise platforms. HR and real estate are getting there too.
We deliberately build for Australian trade and service businesses instead. The triage questions, the after-hours rates, the job systems and the way callers describe a problem are all specific, and that specificity is the reason ours works.
How long does it take to get one running?
Days, not months. We build it from your prices, service area and the rules you give us, you listen to it and tell us what’s wrong, and it goes live on a number you control — usually diverted from your existing one so nothing changes for customers.
The long part is you deciding what it should say when you’re not available. That’s a conversation, not a technical problem.
Stop reading about it. Ring it.
Thirty seconds on the phone tells you more than any page about conversational AI, including this one. It’s a real AI receptionist answering as a real business.
Or have one built as your business
We’ll build a demo answering as your business, free. If it’s not right for you we’ll say so — including if the maths doesn’t work at your call volume yet.