What is an AI chatbot for business, and what makes a custom one different?
An AI chatbot for business is a chat assistant on your website, app or messaging channel that uses a large language model to understand customer questions and reply in natural language. A custom one answers from your own approved content, follows your rules on sensitive topics, and connects to your systems, instead of relying on the model's general knowledge.
That difference is everything. A general-purpose AI model knows a lot about the world and nothing reliable about your delivery times, your returns process or this month's prices. Left to itself, it fills gaps with plausible guesses. A custom chatbot is designed so the model reads the relevant parts of your content first and answers only from them, which is what keeps replies accurate.
For an Australian company, custom also means local detail: prices in AUD, delivery zones that include regional and remote areas, public holiday hours, and refund wording that matches your obligations under the Australian Consumer Law. None of that comes from a model out of the box.
How does a chatbot trained on your own content actually work?
It works by retrieval: your documents are split into short passages and indexed, each customer question is matched to the most relevant passages, and the model writes its answer using only those passages, ideally citing which page it came from. This approach is often called retrieval-augmented generation, or RAG.
“Trained” is a slightly misleading word. In most business chatbots the model is not retrained at all. Instead, your content is kept in a searchable index that we update whenever your policies change. That is cheaper, faster to maintain, and much easier to correct: fix the source page and the next answer is fixed too.
- Collect: FAQ pages, policies, product data, price lists, help articles
- Prepare: remove outdated pages, add dates, split into passages
- Index: store passages with embeddings for meaning-based search
- Retrieve: find the few passages most relevant to each question
- Answer: the model replies from those passages only, with a source link
- Escalate: if nothing relevant is found, offer a person instead of guessing
Which Australian businesses benefit from an AI chatbot for business?
Businesses that answer the same twenty questions all day benefit most: online stores fielding delivery and returns questions, clinics and gyms asked about hours and fees, trades answering service-area questions, education providers handling enrolment queries, and software companies with a busy help desk.
The signal to look for is repetition plus timing. If your inbox fills with near-identical questions, many of them arriving after hours when nobody is there, a well-built chatbot can answer them instantly and pass the rest on with context already captured. If most enquiries are complex, emotional or unique, a chatbot adds little beyond collecting details.
A chatbot also suits businesses with a lot of written material that customers never read: long policy pages, product specifications, course handbooks. The bot becomes a friendlier way into content you already have.
Strong fit
Ecommerce, subscription services, clinics' front desks, education providers, SaaS support, property managers answering tenant FAQs.
Weak fit
Low enquiry volume, highly bespoke sales, or conversations that usually need judgement, empathy or a licensed professional.
What content should an AI chatbot for business be given?
Give it only content you are happy to be quoted on, with dates, and nothing confidential. The chatbot will repeat what it reads, so outdated or contradictory pages become outdated or contradictory answers.
The usual starting set is your FAQ page, delivery and returns policy, terms of service, privacy policy, product or service pages, opening hours and a price list. Internal documents such as staff procedures can be added for an internal bot but kept out of the public one. We run a content review first and send you a list of contradictions to fix, such as two different delivery timeframes on two pages.
Structured data works better than prose for things that change often. Stock, prices and booking availability are best fetched live from your store or booking system at answer time, rather than copied into the index where they go stale. Where an answer depends on a date, such as holiday trading hours, the source should carry that date.
How should a chatbot answer refund and pricing questions under the Australian Consumer Law?
It should give fixed, reviewed answers on refunds, warranties and prices, never improvise them, and never suggest customers have fewer rights than the law gives. The ACCC explains that consumer guarantees cannot be taken away by anything a business says or does, and that telling customers “no refunds” in a way that misleads them about their rights breaks the law.
An AI model that paraphrases your returns policy creatively can easily cross that line, for example by telling someone a faulty item cannot be returned because it was on sale. So we treat refunds, warranties, faults, cancellations and pricing as controlled topics. When a question falls into one, the chatbot uses wording you have approved, links to the full policy, and offers a person for anything involving a specific fault or dispute.
Pricing gets similar care. The bot quotes prices only from live or dated sources, includes the conditions that apply, and never invents discounts. We are not lawyers; your controlled wording should be checked by your own adviser, and we build the chatbot so that updating it is a quick content change rather than a code change.
When and how should an AI chatbot hand over to a human?
A chatbot should hand over whenever it cannot find a relevant source, when the customer asks for a person, when the topic is controlled and specific (a faulty order, a complaint, a billing dispute), or when the conversation shows frustration. The handover should carry the whole transcript so nobody asks the customer to repeat themselves.
We build handoff to fit how your team already works. During business hours, the chat can switch to a live agent in the same window. After hours, the bot collects a name, contact details and a summary, then sends it to your team by email or SMS, and tells the customer honestly when to expect a reply. SMS alerts go through your own gateway account, such as ClickSend or MessageMedia.
The failure to avoid is the bot that refuses to let go. Offering a person should always be one tap away, and the bot should say plainly that it is an automated assistant.
- No relevant source found: offer a person instead of guessing
- Customer types “human”, “person” or “agent”: hand over immediately
- Controlled topic with specifics: collect details, then hand over
- Repeated rephrasing or frustration: apologise and escalate
- Out-of-hours: capture details, give a realistic reply time
What privacy notices does an AI chatbot need in Australia?
Your chatbot should tell users they are talking to an AI, explain briefly what happens to what they type, and link to a privacy policy that describes your use of AI. The OAIC's guidance on commercially available AI products (October 2024) says public-facing AI tools such as chatbots should be clearly identified as such, and that organisations should update their privacy policies and notifications with clear information about their use of AI.
The same guidance recommends, as best practice, not entering personal information, particularly sensitive information, into publicly available AI chatbots. A custom chatbot is different from a staff member pasting data into a public tool, but the principle still shapes our design: we ask for as little personal information as possible, strip it from logs where we can, and use model providers and settings that do not use your conversations for training where that option exists.
Practically, that means a short notice at the start of the chat, a retention period for transcripts that you choose, access to transcripts limited to named staff, and data stored in your own cloud account. Whether the Privacy Act covers your business, and what your notices must say, is for your adviser to confirm.
How much does an AI chatbot for business cost to build?
With BtechWaleTech, a custom AI chatbot for business starts from US$600 and takes 2–4 weeks for a website assistant with one handoff channel. Quotes rise with the number of content sources, live integrations, channels such as WhatsApp, and the testing needed for controlled topics.
The build covers the content review and indexing, the chat interface, retrieval and answer rules, controlled-topic wording, handoff, an admin view of conversations, and a test set run before launch. What it does not cover is the model provider's usage fees, hosting and SMS credits, which are billed straight to your accounts so you can see every dollar.
Quotes from other providers vary widely, often because some include months of platform fees and others exclude testing. Compare them on what is actually built and who holds the accounts. Our starting prices list every service, and a chatbot inside a larger app is quoted as part of a web app from US$900.
What is the cost per conversation of an AI chatbot?
Cost per conversation is the model provider's usage charge for all the text read and written in one chat, plus a small share of hosting and search costs. It is usually small per conversation but adds up with volume, so we measure it from day one and show it in the admin view.
Four things drive it. Model choice: larger models cost more per word than smaller ones, and many support questions do not need the largest. Retrieved context: every passage fed to the model is paid for, so tight retrieval lowers cost and often improves accuracy. Answer length: concise answers are cheaper and better. Conversation length: long chats resend history, so we summarise older turns.
We set a monthly spending cap with the model provider, route simple questions to a smaller model, and cache common answers. We do not quote a per-conversation figure before seeing your content and question mix, because it genuinely depends on both. After a week of real traffic, you will have an actual number.
Should you build a custom AI chatbot or buy a subscription widget?
Buy a subscription widget when your questions are simple, volume is modest and you want something live this week; build a custom chatbot when answers must be tightly controlled, the bot needs to act inside your systems, or you want transcripts and data in your own account.
Subscription tools have improved a lot, and for many small businesses they are the right first step. Their limits show up with controlled topics, live data from your store or booking system, custom handoff to your team's tools, and message caps that climb with plan tiers. A custom build costs more upfront but gives you control over wording, integrations and running costs.
A sensible path for some businesses is to try a widget for a month, read the transcripts to learn what customers actually ask, then commission a custom build shaped by that evidence.
How do you test an AI chatbot before it talks to customers?
Test it against a written set of real customer questions with expected answers, including awkward and adversarial ones, and do not launch until it answers correctly, cites the right source, or hands over appropriately on each. Opinions from a quick demo are not testing.
We build the test set with you from your inbox, reviews and support tickets: usually around a hundred questions covering common FAQs, controlled topics, questions you do not answer, attempts to get discounts, off-topic requests and messages with typos or slang. Each gets a pass or fail. We rerun the whole set whenever content or prompts change.
After launch, testing continues through transcripts. We flag answers where no source was found, where customers asked for a human, and where the rating was negative, then review them weekly with you for the first month.
Where should your AI chatbot live: website, WhatsApp or inside your app?
Put the chatbot where your customers already ask questions. For most Australian businesses that is the website first, then WhatsApp or Messenger if customers message you there, and inside your app if you have one.
One knowledge base can serve every channel. The website widget, a WhatsApp Business number and an in-app assistant can all draw on the same indexed content and controlled-topic rules, so answers stay consistent. Each channel has its own handoff: live chat on the website, the same thread on WhatsApp, a support ticket inside an app.
Leads captured in any channel can flow straight into your pipeline. If you run your own CRM, the custom CRM guide shows how conversations attach to client records.
Does an AI chatbot help SEO or visibility in AI search?
Not directly. Search engines do not rank you higher for having a chatbot, and chat conversations are not indexed content. But the work of preparing content for a chatbot, clear answers, dated policies and structured FAQs, is the same work that helps Google and AI answer engines understand your site.
We often find that a chatbot project exposes gaps: questions customers ask that no page answers. Turning those into proper FAQ entries with structured data helps both the bot and your organic visibility. A fast site matters too, so the chat widget is loaded after the main content to protect Core Web Vitals.
Nobody can guarantee rankings or citations in AI answers. If organic search is a priority, a technical SEO audit is the right companion project.
Working with an AI chatbot team in India from Australia
The project runs in your afternoon and our morning: content review calls and test sessions happen after lunch on the east coast, and written summaries arrive the same day. Queensland's lack of daylight saving keeps the gap fixed at 4.5 hours for Brisbane teams; Perth sits just 2.5 hours ahead of India.
The quote, confidentiality terms and anything else agreed are put in writing before we receive your content; our terms set out the general basis. Invoices come from India in USD, paid by Wise or bank wire. Ask your accountant about GST on imported services.
Week 1
We collect your content, list contradictions for you to fix, agree the controlled topics and draft their wording, and create the model provider and hosting accounts in your name.
Week 2
A working chatbot on a private test page, the first run of the question set, handoff wired to your email or SMS, and your review of every failed answer.
After launch
Weekly transcript reviews for a month, cost-per-conversation reporting, and content updates as policies change, within two months of free maintenance.
Risks and red flags when buying an AI chatbot for business
The main risks are a chatbot that invents answers, promises what you do not offer, hides that it is automated, or sends customer data somewhere you did not agree to. Each can be designed out, so ask any provider how they handle it.
- No test set, just a demo with friendly questions
- Answers without sources or any way to trace them
- No fixed wording for refunds, warranties and prices
- Handoff hidden behind several failed attempts
- Transcripts held only in the provider's account
- Unclear whether conversations are used to train models
- Claims that the bot will replace your support staff entirely
A chatbot should reduce repetitive work and capture enquiries around the clock. It should not be the only way customers can reach you.
Worked example: a Hobart outdoor-gear store adds an AI chatbot
This scenario is invented to illustrate the process. Suppose a Hobart online store selling hiking and camping gear gets dozens of chat and email questions a day: delivery times to mainland states, whether a jacket runs small, how returns work, and whether a tent is in stock.
The content review finds three different delivery timeframes across the site, which the owner fixes first. Sizing notes and product specifications are indexed, while stock levels are fetched live from the store. Returns, faulty items and price matching become controlled topics with wording the owner's adviser has checked. Anything about a specific faulty order goes to a person.
After hours, the bot collects the customer's order number and email and sends a summary to the shared inbox, promising a reply the next business day. The admin view shows which questions the bot could not answer, which leads to four new FAQ entries in the first month. None of this describes a real store; it shows how the pieces fit for a small retailer.
AI chatbot for business: a launch checklist
Use this list before your chatbot goes live. Every unchecked item is a likely source of a bad customer conversation.
- Content reviewed, dated and free of contradictions
- Controlled topics listed, with approved wording for each
- Clear notice that users are chatting with an AI
- Privacy policy updated to describe your use of AI
- Handoff tested in and out of business hours
- A test set of real questions passed and saved for reruns
- Monthly spending cap set with the model provider
- Transcript retention period and staff access agreed
- Owner, not the developer, holds every account
Once the chatbot is working, many teams add a monthly developer to keep improving it; the dedicated React developer page explains that arrangement.