Prepare your knowledge for an AI assistant

Make the source information easier for both your team and a prototype to inspect.

Collect the information people actually use.

Start with service descriptions, frequently asked questions, operating notes and the policies people consult when replying. You do not need to gather every document the business has. Focus on the sources for the task you chose.

Decide which version is approved.

Conflicting documents produce conflicting answers. Mark the current version, identify its owner and remove obsolete copies from the prototype's source set. Note where the assistant should defer to a person instead of offering an answer.

Use clear questions and complete answers.

A short answer that relies on unstated context may work for an experienced employee but be difficult for an assistant to use. Write down the conditions that change the answer. Include the exceptions that matter for customer replies.

Define access boundaries.

Some information can be public; other information is intended only for staff. Keep those sources separate. Start the pilot with the smallest approved source set and decide who is responsible for approving changes.

Test whether answers can be traced.

Ask representative questions and check the source behind each answer. Include questions the source cannot answer. The correct behavior in those cases is to acknowledge the gap and hand the request to a person.

Plan for maintenance.

• Who owns each source?
• How will changes be approved?
• How will outdated answers be noticed?
• Who handles questions outside the source set?

Source quality is an ongoing responsibility. A well-written starting set helps, but it needs a process for staying useful as services and policies change.

Explore Knowledge Assistants or discuss your source information.

Use this guide to prepare your team's discussion. A project's scope depends on your sources, systems, and review requirements.