An AI assistant is only as trustworthy as the information it is allowed to use. If your policies live in old emails, three Google Docs and one employee’s memory, the assistant has no reliable source of truth.
Before building a chatbot or enquiry assistant, build the knowledge base.
Start with information the business is prepared to stand behind
Good knowledge-base candidates include:
- current services;
- opening hours;
- service areas;
- approved pricing ranges where appropriate;
- booking steps;
- refund or cancellation policies;
- common support procedures.
If a policy is still being debated internally, do not give the AI permission to invent the final answer.
Assign an owner to each important source
Information becomes stale because everybody assumes somebody else is maintaining it. For important documents, record who owns the content and when it was last reviewed.
This is especially useful for policies that change with pricing, staffing or regulations.
Prefer clear source material over clever prompts
A long system prompt cannot rescue contradictory business documents. If one page says cancellations require 24 hours’ notice and another says 48, the AI has a source problem, not a prompt problem.
Resolve the contradiction at the content level.
Teach the assistant how to say “I don’t know”
A trustworthy system needs permission to stop. If the answer is not in an approved source, it can ask a clarifying question or route the conversation to a person.
The NIST Generative AI Profile is useful background for managing risks such as confabulation and unreliable output.
Limit access to what the use case needs
A customer FAQ assistant does not need your payroll folder. Keep the knowledge scope narrow and separate public business information from confidential internal material.
The OWASP GenAI Top 10 is a useful reference when evaluating access and data exposure risks.
Test with real customer language
People rarely ask questions using your internal policy headings. Test variations, misspellings and incomplete questions. Check whether the assistant retrieves the right source and whether it knows when to escalate.
Our Customer Enquiry Automation service treats the knowledge base as part of the product, not as an afterthought.
Review the knowledge base as the business changes
If your services, hours or policies change, update the source before expecting the assistant to behave differently. A reliable AI system is partly a content-maintenance system.