⏱️ The Essentials in 3 Minutes |
🧠 Understand the Value of Choosing Documents Carefully
Every document integrated into the knowledge base influences what the AI can answer, and how. An outdated, redundant, or audience-inappropriate document can generate incorrect or off-topic responses, even if your base otherwise contains quality content.
Choosing documents methodically ensures the AI becomes a genuine skills assistant for your teams, rather than a source of confusion. It also ensures your knowledge base-building effort is sustainable: a small, well-maintained base always outperforms a large, unmaintained one.
🎯 Define Your Learners' Needs
Before importing anything, clarify the situations in which the AI needs to be useful.
Recurring needs:
On which topics do your collaborators come to you most often (questions in meetings, messages, emails)?
When someone joins your team, what do they lose time on during the first few weeks?
Grey areas and edge cases:
Are there recurring decisions where your collaborators do not always know which rule to apply?
Are there nuances or exceptions that only long-standing team members know?
Are there topics where practices vary from person to person, due to a lack of a common reference?
🔍 Identify Documents That Address These Needs
For each identified need, ask yourself:
On whether the document exists:
Is there already reliable information addressing this need somewhere (procedure, guide, internal policy, FAQ, presentation)?
Is there a reference answer given verbally but never written down? If so, it should be documented before being integrated.
On the access scope:
Does this information concern all your learners, or only a specific population (managers, a team, a particular training space)?
Is there a risk in making this information accessible to everyone (sensitive data, HR information, content reserved for certain roles)?
If the information only concerns a subset of your learners, restrict the document's access rights at import time.
✅ Check Document Quality Before Ingesting
A poor document in the base can degrade response quality just as much as missing information.
Reliability:
Is the document up to date and does it reflect current practices?
Is it understandable on its own, without verbal explanation or additional context?
Is there another document on the same topic in the base? If so, are the two consistent? Can the new document be restricted to additional information only, to avoid duplication?
Sustainability:
Is the content stable over time, or likely to change frequently (pricing, contacts, ongoing decisions)?
If the document evolves, who will be responsible for updating it in the base?
🧪 Apply the 3-Question Test
Before any import, check these three criteria:
A learner has already needed this information, or will likely need it.
The information is correct, up to date, and self-contained: it stands on its own.
Someone is responsible for maintaining it: the base degrades if no one is in charge.
If you answer "yes" to all three: import. Otherwise, take the time to correct before integrating.
Keywords: document selection, ingestion, knowledge base, Didask AI, quality, access rights, document strategy.