What determines AI assistant cost?

What you are really paying for

A useful AI assistant is not just a chat bubble on a website. For a small business, the real work is deciding what the assistant is allowed to know, what it should say, when it should stop, and how it hands the customer to a person. That planning affects setup effort more than many owners expect.

Start with the bottleneck, not the tool. If customers cannot find your opening hours, service area or booking link, clearer website content may be the first fix. If the bottleneck is repeated questions after hours, missed enquiries, or staff spending time qualifying simple requests, an AI assistant may be worth scoping. Measure the current issue first: common questions, missed calls, enquiry quality, response time and the number of conversations that need a staff member.

Sources: [1]

Off-the-shelf or tailored

Off-the-shelf AI tools can be quicker to try and easier to start. They may suit internal drafting, brainstorming, meeting summaries or simple experiments where the risk is low and your staff can check the result before anyone outside the business sees it. This is often the right place to begin if you are still learning what AI can and cannot do for your business.

A tailored assistant is different. It should reflect your services, exclusions, tone, opening hours, service areas, policies and handoff rules. That can make it more useful for customer enquiries, but it also means more preparation, testing and ongoing care. Australian Government guidance notes that custom AI solutions can be more flexible, but usually need high-quality, well organised data and ongoing support and maintenance.

Sources: [1]

The main cost drivers to scope

Approved knowledge preparation is usually the first major cost driver. Someone needs to turn your existing website, FAQs, service descriptions, policies, booking rules and staff knowledge into approved information the assistant can use. Messy information creates extra work because the assistant needs clear boundaries, not vague instructions such as answer customer questions.

Channels are the next driver. A website-only assistant is simpler to define than an assistant that also works through voice or other customer contact points. Each channel needs its own greeting, fallback wording, handoff process and test conversations. Extra channels can be useful, but they should match the way customers already contact you.

Tool integration also changes the scope. There is a big difference between answering common questions, collecting enquiry details, sending a customer to a form, connecting with a booking system, or handing off to a secure payment provider. Taking bookings requires a compatible booking integration; a chatbot alone cannot confirm live availability unless it can safely connect to the system that holds the calendar and booking rules.

Included usage matters too. Higher enquiry volume, longer conversations, voice use and more complex answers can all affect how usage should be planned. Rather than guessing, ask what is included, what happens if usage grows, and how extra work or usage will be approved before anything changes.

Sources: [1] [6]

Two fictional scopes compared

Fictional scope A: a local service business wants a website Q&A assistant. It answers from approved public business information, explains what the business does and does not do, asks one or two qualifying questions, then hands the enquiry to the team. It does not take payments, make promises, give advice outside the approved information, or confirm bookings. This scope is easier to brief because the assistant has fewer jobs and fewer systems to touch.

The trade-off is that scope A may not remove much admin if the real bottleneck happens after the enquiry arrives. It can still improve customer guidance, but staff may continue to manage bookings, follow-up and detailed questions manually. That can be sensible if the business wants a cautious starting point and a clear way to test customer behaviour before adding more complexity.

Fictional scope B: a growing allied health clinic wants an enquiry assistant that answers common service questions, clarifies whether the person is a new or returning client, works on the website and phone, and sends people towards a compatible booking or follow-up process. This needs more approved rules, more testing, clearer privacy boundaries and stronger human handoff. It may be more useful, but it carries more implementation effort and more ways for the experience to fail if the systems are not ready.

Practical buying checklist

Before asking for a proposal, write down the decisions below. You do not need technical language; you need practical boundaries. A clear brief helps a supplier recommend the right starting point instead of pushing a chatbot where a content fix, form, booking tool or internal workflow would solve the problem better.

• What customer problem are we trying to reduce: missed calls, repeated questions, poor enquiry quality, slow replies or something else?
• What are the top 20 questions customers ask before they contact us?
• Which answers are already approved, current and safe to use?
• What must the assistant never answer, guess or promise?
• What personal, financial, health or sensitive information must not be collected in chat?
• Which channels are in scope now: website, phone or another customer contact point?
• Should it only answer questions, or also collect enquiry details, qualify leads or hand off to another system?
• If bookings are involved, is there a compatible booking system with accurate availability and rules?
• Who in the business will check test conversations before launch?
• What will we measure after launch: conversation volume, handoffs, unanswered questions, booking clicks, enquiry quality or staff time spent triaging?
• Who will update the assistant when services, hours, locations, staff, policies or offers change?
• What is included in ongoing care, and what counts as extra scope?

Risk, testing and ongoing care

Customer-facing AI needs more care than private drafting because customers may rely on what it says. Official cyber guidance for small businesses warns about data leaks, privacy breaches, unreliable outputs, prompt injection and supply chain dependencies. In plain English, this means an assistant can expose information it should not, make up a confident answer, be manipulated by a user, or depend on a third-party service with its own risks.

Good implementation limits what the assistant collects, uses approved information, keeps sensitive matters out of chat where possible, and gives customers a clear path to a person. High-risk advice, complaints, legal, medical, financial or safety-related questions should have human oversight or be handed off rather than answered freely.

Ongoing care is not optional if the assistant represents your business. Services change, staff change, booking rules change and old answers become wrong. Test before launch, monitor real questions, review failed conversations and update the knowledge base when your business changes. Also decide who can approve changes, because letting anyone add instructions can create inconsistent or unsafe answers.

Sources: [2] [4] [5]

When professional help is worthwhile

Professional help is most useful when the assistant will speak to customers, use business-specific rules, work across more than one channel, or connect to tools such as booking, enquiry follow-up or secure payment handoff. It is also useful when the business owner does not have time to prepare approved knowledge, run test conversations, check data handling or maintain the assistant after launch.

Business Growth Clinic’s documented service is relevant because tailored AI chat sits within the website and ongoing care package, rather than being presented as a published standalone AI price. The assistant is scoped around approved business information, practical handoff and compatible integrations where they make sense. That matters because the right answer may be a clearer page, a better enquiry path, a simple assistant, or a more involved automation.

If you are comparing options, do not start by asking what does an AI assistant cost? Start with what should this assistant do, what should it never do, what systems must it touch, and how will we know it is working? If you want help turning those answers into a clear scope, Business Growth Clinic’s AI & automation service at /services/ai-automation is the natural next place to explore.

Sources: [6]