01 / ACCESS
AI can remove the friction from finding an operational answer.
IskayPet built an internal knowledge assistant over hundreds of SharePoint documents in four languages. In an AWS case study, the company reported that average search time fell from three to five minutes to about seven seconds. The useful design choices matter more than the headline result: answers cite their sources, respect document permissions, and say when the knowledge base does not contain an answer. For a small business, the first win is often not generating new material. It is making existing policies, procedures, and job knowledge easier to find without losing the source or its access controls.
02 / FRESHNESS
Faster access does not fix stale source material.
The IskayPet system uses incremental synchronization so updated documents become available to the assistant. Google Cloud separately describes why this matters: business definitions and rules change, while static documentation falls behind. An assistant connected to old material can deliver an outdated answer more confidently and more consistently than a person searching manually. Connecting AI to a document library is not a one-time implementation. Every trusted source needs an owner, a review cadence, and a clear way to retire obsolete versions.
03 / OWNERSHIP
The business still has to decide what counts as authoritative.
Google Cloud's approach stages proposed knowledge changes with citations and confidence information so a responsible steward can review them. The AWS example limits each assistant to relevant knowledge and turns employee feedback into an evaluation set. Both point to the same operating principle: software can surface gaps and conflicts, but an accountable person must approve the answer the business will rely on. The knowledge assistant is not the source of truth. It is the interface to the source of truth. Assigning authority before scaling access prevents a convenient tool from becoming a faster rumor system.
THE CJC VIEW
Implementation beats experimentation.
A knowledge assistant does not create operational discipline. It reveals whether that discipline already exists.
Better search amplifies whatever is in the source library—accurate guidance, duplicate files, or outdated procedures. The practical sequence is to establish ownership first, then make the material easier to use with AI.
PRACTICAL NEXT STEP
Audit the ten documents an AI assistant should trust
- Choose one workflow where employees repeatedly ask the same internal questions.
- List the ten documents most likely to contain the answers, then identify duplicates, conflicts, and obsolete versions.
- For each trusted document, record an owner, last-reviewed date, next review date, and one authoritative location.
- Define what the assistant should do when sources conflict, are expired, or do not contain an answer: cite the source, state the uncertainty, and escalate to the owner.
- Test ten real employee questions, record weak answers, and use those misses to improve the source material before expanding the system.
Reliable AI answers begin with a maintained source system, not a bigger prompt. Give important documents an owner and a review date before asking an assistant to make them easier to find.