Every practice below protects one rule: capture should feel like nothing, not like a task. If a practice makes someone stop and think “ugh, this again,” it is the wrong practice, no matter how complete the knowledge it produces.
The rule underneath everything else: experienced people already say their judgment out loud, constantly, in the normal course of doing their job. The work is noticing it, not asking for it.
Slack answers, ticket resolutions, call transcripts, and CRM notes already contain the reasoning behind most nonstandard decisions. Pull from these first. This should account for most of your library before a single person is ever interrupted.
Some judgment never gets said out loud anywhere—it is too fast, or too instinctive, to show up in a transcript. That is the only category worth an active ask, and it should be a short list: decisions with both high frequency and high variance across who is handling them.
“Walk me through your process” produces a generic, sanitized answer. “Tell me about that renewal call an hour ago” produces the real thing, because it is a specific memory instead of an abstract summary. Always tie the ask to a real, recent event.
Narrative paragraphs cannot be reasoned over by an agent. A decision card can: trigger, cue, move, why, break. Five short fields, twenty seconds to scan, structured enough to be retrieved when a similar call comes up again.
Requiring an explicit approval on every card is still a task, even at ten seconds. Publish automatically and let the contributor flag a problem if there is one. Silence should mean agreement, not require it.
The single biggest driver of continued participation is visible reuse—seeing that a card you gave got pulled by a teammate or an agent, with your name on it. A library nobody sees getting used is a library people quietly stop feeding.
Do not run this against the whole organization evenly. Identify the go-to people—the ones others route exceptions to, whose absence stalls something—and concentrate effort there. Everyone else contributes passively or not at all.
Most cards should come from mining, not asking. If active requests are the majority, the passive layer is not built yet.
Hours, not weeks. Judgment is freshest right after it is exercised.
This should stay near 100 percent because the ask is rare and specific enough that it is not worth ignoring.
Rising reuse is the real signal that the library has value—not the count of cards captured.
Business Knowledge × AQ × AI = Business Value
These practices exist to protect the AQ term in that equation. Every one is a guard against the thing that kills capture efforts before they produce anything: turning someone’s expertise into homework.