Getting an AI assistant to answer questions takes an afternoon. Getting your team to still be using it three months later is the hard part.
Answer from your own documents, and show the source
Ask a general-purpose model about your leave policy and it will answer confidently, and possibly wrongly. An internal assistant should search your own material first (policies, process guides, product manuals, old support tickets) and answer from what it finds.
It should also link to where the answer came from. That link is what lets people trust it, and it shows you quickly when a document is out of date.
One thing that's easy to miss: the assistant has to respect existing permissions. If someone can't open the HR folder, the assistant shouldn't quote it to them. And if your tickets contain customers' personal information, you're in POPIA territory, so decide upfront what gets indexed.
Test it on real questions before launch
Collect 50 questions people asked in the last month, from email, Teams or the helpdesk. Run them through the assistant and have someone who knows the answers mark each one right, wrong or half right.
If it gets fewer than about four in five right, hold the launch. The fix is usually in the source documents rather than the prompt: most wrong answers trace back to missing or contradictory documentation.