Perspective
With business AI tools, start with your data
An AI tool that works with an organisation's own knowledge is only as good as that knowledge. Choosing a tool before scattered documents, outdated versions and unclear permissions are sorted out is a premature step.
Verimel2 min read
AI-assisted search and query tools can produce answers from an organisation’s documents, records and correspondence. Set up properly, they bring real convenience: procedures, contract clauses or technical notes are found in seconds instead of minutes. But the success of these tools depends less on the model itself than on the order of the information it draws on.
Scattered information, unreliable answers
If three versions of the same procedure sit in different folders, the tool cannot know which one to rely on. An outdated document carries the same weight as the current one. The result can be a fluent but wrong answer. So the first task is not choosing a tool but sorting the information:
- Which documents are current and valid?
- Have older versions been moved to an archive?
- Are each document’s owner and last update date known?
Access permissions
An AI tool should never go beyond what the user is allowed to see. Information from a folder an employee cannot normally open should not appear in an answer through the tool. This requires the existing access setup to work correctly with the tool. Shares left open to “everyone” over the years can suddenly become visible in unexpected ways once an AI tool is connected.
A small, measurable start
A good start is an area with clear limits, for example only HR procedures or only technical support notes. In that area:
- Source documents are sorted and updated.
- The tool is connected only to those sources.
- The accuracy of answers is checked against frequently asked questions.
- Users are shown which document each answer is based on.
An answer that shows its source makes it easier for users to verify and builds trust in the tool.
Where the data is processed
Where the tool processes data, whether the provider uses that data to train models and how long records are kept are worth clarifying during selection to avoid surprises later. For documents containing personal data or trade secrets, it makes sense to go through these topics with security and legal advisers.
The result
An AI tool does not fix an organisation’s information practices; it makes the existing ones more visible. An orderly, current and correctly permissioned knowledge base gives better results whatever the tool. That is why the first investment should often go not into software but into the information itself.
