Practical AI · 4 min read
Before you add AI chat, decide what it is allowed to know.
The useful question is not whether a chatbot can answer. It is whether it can answer from the right information, recognize a limit and leave your team with something they can act on.
Give it a maintained source of truth
Start with current service descriptions, hours, locations, contact routes and approved FAQs. Identify who owns each fact and how it changes. Conflicting prices in two documents create a business problem that a longer prompt cannot reliably fix.
Separate public answers from private records
A public visitor can ask about opening hours. An order balance or customer account needs verified identity and authorization before retrieval. Knowing an email address is not, by itself, permission to view everything attached to it.
Define what it must not promise
List commitments that require a person or a confirmed system response: pricing exceptions, delivery dates, refunds and appointments. Give the assistant a useful fallback when it cannot answer. A plain explanation and a handoff are better than a confident guess.
Make the handoff part of the design
Choose the destination for a new enquiry, the minimum information needed and the person responsible. Preserve relevant conversation context. Tell visitors when a human response is expected only if the business can support that expectation.
Review the difficult conversations
Test ambiguous questions, outdated information and requests outside the agreed scope. Revisit unanswered questions after launch. Track whether a handoff reached someone, not merely how many chat messages were generated.
Further reading
Microsoft: RAG design and evaluationYour next AI project starts here
Bring the challenge.
Let’s build the AI.
Show us the task, the information and the tools involved. We’ll help you shape a useful first project.