AI chatbot go-live checklist

Start with the enquiry bottleneck, not the tool

Before you add chat to your website, be clear about the business problem. A chatbot may help if you receive repeated questions after hours, miss basic enquiry details, or spend time answering the same service-area, opening-hours or booking-process questions. If the real problem is that your website does not explain what you do, your prices are confusing, or your contact form is hard to find, fix that first. AI is useful when it supports a defined process, not when it is used to cover unclear offers.

Australian Government guidance recommends identifying the problems AI should solve, starting small, testing tools and involving staff. That is a good standard for a customer enquiry chatbot. Treat the first launch as a controlled trial: one website, one enquiry pathway and a short list of approved answers. If it works, you can expand its role later; if it creates confusion, you can tighten the rules before more customers rely on it.

Sources: [2]

Define the safe job for the chatbot

A safe small-business chatbot has a narrow job. It can answer common questions from approved business information, clarify what the customer needs, point to the right next step and collect only the minimum details needed for a human follow-up. It should not make judgement calls, provide professional advice, invent custom prices, promise outcomes or argue with customers. If the answer is not in its approved information, it should say so and offer a handoff.

Bookings need special care. A chatbot alone cannot confirm real availability unless it is connected to a compatible booking integration that can check open times and create or update the booking. Without that integration, the chatbot can explain the booking process, collect a preferred time or direct the customer to the booking page, but it should not say the appointment is confirmed. This is an implementation boundary to decide before launch, not something to discover through customer complaints.

Sources: [1] [2] [6]

Go-live checklist for customer enquiry chat

Use this checklist before launch. It is designed for a practical first version of a website enquiry assistant, not a complex enterprise system. If any item is unclear, launch with limits or pause until the rule is decided.

• List the approved knowledge the chatbot can use: services, service areas, hours, booking steps, turnaround expectations and escalation paths.
• Remove or rewrite anything outdated, vague or promotional that could cause the chatbot to overpromise.
• Decide what the chatbot may collect, such as name, contact details, suburb, preferred service and a short enquiry summary.
• Block collection of passwords, payment card details, unnecessary identity documents, confidential business information and sensitive personal details.
• Set handoff rules for complaints, urgent issues, cancellations, refunds, custom quotes, legal or financial questions and uncertain answers.
• Make it clear to customers when they are talking to an AI assistant and how they can reach a person.
• Test common questions, unusual wording, spelling mistakes and questions outside your services before launch.
• Test misleading prompts, such as attempts to get private instructions, hidden data or answers outside the approved knowledge.
• Confirm whether customer chat data is stored, where it is processed, who can access it and whether it may be used to train AI models.
• Choose who in the business reviews chat transcripts, updates approved information and handles escalated enquiries.
• Create a stop rule for serious issues, such as wrong advice, privacy concerns, repeated failed handoffs or customer complaints.
• Record the chatbot in your AI register or internal systems list, including its purpose, owner, data use, risks and review date.

Sources: [1] [2] [5]

Worked example: after-hours enquiries for a local service business

Hypothetical example: a mobile dog grooming business in suburban Brisbane receives many enquiries at night. The owner wants the website chat to reduce back-and-forth and stop customers waiting until the next morning for simple answers. The safe chatbot job is to explain service areas, describe standard services, outline the booking request process and collect a name, phone number, suburb, dog size and preferred service. It also asks whether the customer wants a call back or a booking link.

The chatbot is not allowed to take card details, diagnose skin issues, handle aggressive-dog safety decisions, offer custom discounts or promise a specific appointment time. If the booking calendar is not integrated, it says that the team will confirm availability. If a compatible booking integration is added later, the scope can change so the chatbot can help with live booking steps inside the limits of that system. Until then, it is an enquiry assistant, not a confirmed-booking engine.

Go, limit or pause before launch

Use a simple decision rule once the checklist is complete. Launch if the chatbot only uses approved information, avoids sensitive data, has clear handoff rules, tells customers it is an AI assistant and has passed your basic test questions. This is usually the right path when the assistant is handling general enquiries and routing customers to your team.

Launch with limits if one part is still uncertain but can be safely restricted. For example, you might allow general service questions but disable booking claims, quote estimates or detailed intake questions. Pause if you cannot explain what data is collected, where it is stored, whether it may be used to train AI models, who reviews failures or how customers reach a person. A paused launch is not a failed project; it is a sign that the rules need to be clearer before customers use it.

Sources: [1] [2] [5]

What to measure after launch

The first month should tell you whether the chatbot is reducing friction or creating extra work. Measure practical signals, not vanity numbers. Useful measures include the number of conversations, how many became enquiries, how many required human handoff, which questions failed, and whether customers submitted enough detail for the team to respond. Also check whether staff trust the summaries or need to re-ask the same questions.

Review risk signals as well. Look for customers sharing information they should not share, the chatbot giving answers outside its approved scope, complaints about confusing responses, and any signs of prompt manipulation. AI outputs can be wrong, incomplete or biased, so important answers need human oversight. Make one person responsible for reviewing transcripts and updating the approved information when services, hours, locations or policies change.

Sources: [2] [5]

When you need help implementing it

You may be able to handle a simple chatbot yourself if your services are straightforward, your website content is current and your team can review chat transcripts regularly. Help becomes more useful when the chatbot must match your website content, approved service rules, enquiry form, phone workflow or booking process. It is also sensible to get help if your enquiries include sensitive personal details, complaints, custom quotes, regulated advice or multiple staff handoff points.

Business Growth Clinic’s AI & automation work is suited to bounded customer enquiry assistants: grounded in business information you approve, scoped around your systems and designed with a handoff to your team. The aim is not to make AI the first answer to every problem. It is to decide whether a chatbot fits the observed bottleneck, then build a practical flow with clear limits if it does.

Sources: [6]