Model choice gets the attention. Knowledge design determines the result. When an agent gives a confident wrong answer about a policy, a price or a procedure, the cause is almost always upstream: the current version was not in scope, two sources disagreed, or the agent could not tell which document applied to which case.
Four questions for every source
Before a source enters an AKE deployment we ask: Is it authoritative, and who says so? Is it current, and how do we know when it changes? Who is allowed to see it, and does the agent inherit those limits? And what context does a reader need to apply it correctly: the jurisdiction, the product line, the date?
The answers become the knowledge-source and permission map, which is the first deliverable of the Design stage. It is a short document. It is also the one clients come back to most often, because it is the first time anyone has written down where the organisation’s working knowledge actually lives.
Terminology is knowledge too
Across our offices in Taipei, Kuala Lumpur, Melbourne and Auckland we work in organisations whose knowledge spans Mandarin, Malay, Japanese and several varieties of English. The same concept has different names in different plants, teams and countries. A deployment that ignores this answers correctly in one office and wrongly in the next. Capturing terminology, with its owners, is part of the foundation, not a nice-to-have.
Good enough for this workflow
None of this requires a multi-year data programme. The foundation is scoped to one workflow at a time: the sources it needs, fixed to the standard it needs. Perfection everywhere is not the goal. Traceable, current, permitted information for the task at hand is.