Staff spend a surprising amount of time searching for answers that already exist somewhere: in a policy, a procedure manual, a previous report or a shared folder. A knowledge assistant lets people ask questions in plain language and receive answers grounded in the organisation’s own documents.
How these assistants work
Most knowledge assistants use an approach called retrieval-augmented generation:
- Your documents are split into passages and indexed so they can be searched by meaning.
- When someone asks a question, the most relevant passages are retrieved.
- A language model writes an answer using those passages, and ideally cites them.
Because answers are based on retrieved sources, the assistant can work with up-to-date internal information without the model being retrained.
Good sources make good answers
The assistant can only be as reliable as the documents behind it. Before building, review the content:
- Remove outdated and duplicate versions.
- Identify an owner for each important document.
- Make sure documents are readable as text, not only as scanned images.
Always show sources
Answers should link to the passages they came from. Citations let users verify information, build trust and spot when the assistant has misunderstood a question.
Respect permissions
Not everyone should see everything. The assistant must only retrieve documents the person asking is allowed to read. This needs to be designed in from the start, using the same access rules as the original systems.
Handle uncertainty honestly
Configure the assistant to say when it cannot find a relevant source rather than guessing. An honest “I could not find this in the available documents” is far better than a confident wrong answer.
Evaluate with real questions
Collect real questions from staff and check the answers against the correct information. Track how often the assistant answers correctly, cites the right source or declines appropriately. Repeat this testing whenever documents or settings change.
Keep it maintained
Documents change. Plan how new and updated documents will be indexed, how retired ones will be removed and who reviews the assistant’s performance.
A well-built knowledge assistant does not replace expertise. It helps people find the right information faster and spend more time applying it.
