A useful AI chatbot brief describes a job, its information sources and the point where a person takes over. “Add AI to our website” leaves too many decisions unresolved. A polished demonstration is not enough to show that a system will work with your customers and business processes.
Before requesting a proposal, collect the following information. You do not need a complete technical specification, but a clear business brief makes discovery more useful.
Define the job and the users
Choose the first task the chatbot should support. Examples include answering questions from an approved service guide, helping a visitor find a relevant page or collecting a short enquiry for a human response.
Name the users and the limits. A public website assistant and an internal knowledge tool have different access requirements. Write down what the assistant must not decide or promise. Keep approvals, sensitive decisions and unusual requests with an appropriate person.
Identify approved source material
List the documents or pages that can answer the questions. Give each source an owner and a review process. If a price list or service description changes, the system needs a reliable way to receive the approved update.
Do not assume that a model knows your current policies. Include examples of contradictory or missing information and decide how the assistant should respond. Asking a clarifying question or directing someone to your team may be the correct outcome.
A website AI chatbot and a customer-support AI workflow should be scoped around the actual information and response process available.
Describe the human handoff
Specify when the conversation moves to a person: uncertainty, a complaint, a request outside the service, a change to an account or an action requiring approval. Explain what context is passed along and where it is stored.
Test the handoff as part of the product. A message saying “our team will contact you” needs a working route, a responsible person and a process for reviewing the request. Avoid collecting information simply because the chat interface makes it easy.
Write test conversations before building
Create a small set of representative questions with the response you would accept. Include an ordinary enquiry, a vague request, an outdated fact, a question outside scope and a request the assistant should refuse or escalate.
Agree how each example will be reviewed. Check factual accuracy, use of approved material, appropriate escalation and whether the visitor understands the next step. Repeat the evaluation after changing the instructions or knowledge sources.
A question such as “Can you guarantee my campaign will double sales?” should not produce an invented commitment. The assistant should reflect the business’s actual position and direct the prospect to a scoped discussion.
Scope integrations as actions with permissions
If the system creates CRM records, sends notifications or updates software, define each action separately. State which fields may be written, when approval is required and what happens if the action fails.
An integrated CRM workflow needs a way to detect duplicate records, review exceptions and maintain access. Start with the smallest useful set of actions rather than connecting every system at once.
Plan ongoing ownership
Choose who maintains the source material, reviews errors and handles changes after launch. Identify the external services and usage charges that may be required. An agency’s free initial discussion does not make a production model, hosting or third-party service free.
For a remote engagement with a UK business, we agree the task, approved information, collaboration and human review during discovery. See our AI and automation support for UK businesses for the starting questions.
What to send with an enquiry
Share the task, intended users, a sample of non-sensitive source material, the systems involved and two or three difficult example questions. Do not send passwords or private customer records through the initial enquiry form.
Keari Global can help turn that brief into a defined AI consulting or development scope. Discuss your project with the business problem first.