01Start with the job, not the chatbot
A chatbot is worth building when a well-defined set of questions or tasks arrives often enough to justify automation, and the correct answers already exist in your policies, help content or systems. Write down the top questions, where today’s correct answers live, and what a good outcome looks like: a resolved question, a booked appointment, a qualified lead or a hand-off to a person.
02Choose the kind of assistant
Most business chatbots fall into three groups, and the difference decides both the build effort and the risk.
- Support and FAQ assistants answer from approved content, usually by retrieving passages from your documents.
- Transactional assistants look up or change records — an order status, a booking, an address — through your systems’ APIs.
- AI agents plan several steps and call tools on their own, so they need the tightest permissions and approvals. Compare the two in AI agent vs chatbot.
03Ground every answer in content you own
The risk of invented answers falls when the bot answers only from a defined knowledge base and says so when it cannot help. Decide who owns that content, how it is kept current, and what the bot must refuse to answer. In February 2024, British Columbia’s Civil Resolution Tribunal held Air Canada responsible after its website chatbot told a customer he could apply for a bereavement fare retroactively, contrary to the airline’s policy; the tribunal rejected the airline’s argument that it could not be held liable for information its chatbot provided (Moffatt v. Air Canada, 2024 BCCRT 149). If a bot speaks for your business, its answers are your answers.
04Plan the hand-off to people
Every chatbot needs a clear way out: when its confidence is low, when the topic is sensitive, or simply when the customer asks. Decide which channel takes over, what context the person receives, and when that hand-off is available.
05Test before your customers do
Build a test set from real questions, including awkward and adversarial ones, and score the bot before launch and after every change. In March 2024, The Markup reported that New York City’s MyCity chatbot gave business owners answers that would break the law, such as telling employers they could take a share of workers’ tips. Public, official-looking bots need the strictest testing. The method is set out in how to evaluate an AI application before launch.
06Questions to ask an AI chatbot development company
Use these to compare proposals on substance rather than on demos:
- Which content and systems will the bot use, and how is it kept from answering outside them?
- How will you measure answer quality, and what score must it reach before launch?
- How does hand-off to a person work, and what will that person see?
- Where are conversations stored, for how long, and who can access them?
- What will running costs depend on — model usage, hosting, monitoring — and how are they capped?
07What drives the build effort
Effort rises with the number of systems the bot must act on, the volume and quality of content to prepare, the languages and channels it must support — web, WhatsApp or an app — and the evaluation and monitoring your risk level needs. The AI application development cost guide explains each driver, and choosing an AI application and AI agent development partner covers what a good proposal includes.
Editorial approach
This guide offers practical planning questions. It is not a guarantee of delivery outcomes. Adapt the checklist to your project and validate assumptions with the proposed team.