The knowledge base has 4,200 articles. Search returns three of them, none from after 2022, all contradicting each other. The team’s response is to file more tickets instead of trusting the KB. Knowledge Management does not fail loudly — it fails quietly, one stale article at a time, until users stop checking. The discipline below is what separates a trusted KB from a digital landfill.
Lifecycle: Draft to Review to Publish to Retire
Every article needs an explicit lifecycle. Drafts are not visible to general users. Reviews have named owners with SLAs. Published articles have explicit expiration dates that the platform enforces. Retirement removes the article from search and marks it for either archive (preserved for audit) or delete (after legal hold expires). Skipping any state — usually “Retire” — is how the KB grows past usable size.
State machine:
Draft -> In Review (assigned to category owner)
In Review -> Published (with expiration)
Published -> Pending Re-certification (60 days before expiration)
Pending -> Re-certified OR Retired
Retired -> Archived OR Deleted (per retention policy)Ownership per Category
Every knowledge category has a named owner. Without ownership, nobody reviews. A five-year-old article at the top of search results is worse than no article — it produces wrong action with confidence. The category owner’s responsibilities include reviewing pending re-certifications, approving new drafts, and signing off on retirement. Categories without an active owner enter a “stewardship gap” state and new drafts are blocked.
Feedback Loops
Enable user feedback on every article (helpful, not helpful, free text). Route negative feedback to the article owner weekly with the original article context, the user’s free text, and the searched query that surfaced the article. Articles accumulating 10 or more “not helpful” votes within 90 days flag for review. Without this loop, the KB does not know it is failing until it has failed for a long time.
// Scheduled aggregator for negative feedback
var ag = new GlideAggregate('kb_feedback');
ag.addQuery('helpful', false);
ag.addQuery('sys_created_on', '>', gs.daysAgo(90));
ag.addAggregate('COUNT', 'article');
ag.groupBy('article');
ag.addHaving('COUNT', '>=', 10);
ag.query();Search Analytics
What do users search for that returns nothing? That is your content gap. Review zero-result searches monthly and commission articles for recurring queries. The kb_search_log table holds the data; a Performance Analytics indicator on top of it makes the trend visible. Recurring zero-result searches for the same term means the article either does not exist or is not surfacing — both problems worth investigating.
Retention Policy
By default, articles live forever. Override that. Set maximum age per category — product documentation might be two years, troubleshooting runbooks one year, policy documents three years. Auto-expire when the limit hits and require re-certification by the owner to extend. Categories without explicit retention default to a conservative one-year limit; this forces the policy decision rather than allowing eternal life by default.
Common Failure Modes
Articles re-certified by the owner without actually reading them — re-certification should be a sample-driven audit, not a checkbox. Negative feedback ignored because the owner has too many articles — the volume problem is itself a signal that the category is over-broad and needs splitting. Search returning the article but ranking it last — boost recently-updated articles in the search relevance config, otherwise the freshly-revised article competes with the stale one users keep clicking.
What Changed in 2026
Now Assist’s knowledge generation (where licensed) drafts articles from resolved incidents. The drafts go through the same lifecycle as human-authored articles — review, publish, expiration. Treat them as productivity tools for the article author, not autonomous publishers; the human review step is the quality gate that keeps generation from polluting the KB.
What to do this week: query kb_knowledge for articles published more than 18 months ago with no update in the last 12 — that count is your re-certification backlog and the most urgent risk to KB trust.
