Implementation / Northbound Frontier
Connecting AI to company knowledge
Useful answers depend on relevant sources, appropriate access and a clear way to say “I don’t know”.
Begin with the source material
Identify which documents are authoritative and who maintains them. Duplicates, old policies and unclear ownership can undermine an answer before a model is involved. Start with a small, curated collection.
Preserve access boundaries
A knowledge workflow should respect the permissions of the person asking the question. Connecting a document store is only one part of the work; retrieval and answer generation need to preserve the same access rules.
Make sources inspectable
Include references that let people inspect the passages used in an answer. Citations support review, but do not by themselves prove an answer is correct. Test whether the answer is supported by the cited material.
Test unanswered questions
Include questions that the collection cannot answer. A clear statement that information is missing is often more useful than a plausible guess. Decide whether the next step is to ask a colleague, find another source or revise the question.