Why do knowledge bases decay?
Knowledge bases decay because writing and updating articles is nobody's primary job. Support agents resolve tickets; product teams ship changes; the article that would have prevented the next fifty tickets sits in a backlog. Meanwhile, older articles quietly go wrong as the product changes.
The decay is structural, and it compounds: every gap generates tickets, and every ticket takes time that could have closed the gap. A Digital FTE whose job includes noticing the gaps breaks the cycle.
How does the Digital FTE propose articles?
The Digital FTE proposes an article when the same question escalates repeatedly because no source covers it. The draft is written from how the team actually resolved those tickets, with the resolution steps, the conditions that applied, and any product references cited, and it arrives in Approvals with the tickets that prompted it.
The support lead approves, edits, or sends it back with a reason. Once published, the next ticket of that type resolves at first line with a citation to the new article, which is the whole point.
How does it find stale or wrong articles?
It finds stale articles from evidence in the tickets. When customers who read an article still open a ticket about the same thing, the article is being misread or is wrong. When two articles give different answers to the same question, they contradict. When a product change is announced, articles referencing the old behaviour are candidates for review.
Each flag arrives with the evidence: the tickets, the conflicting passages, or the product change. The team decides what to do; the FTE can draft the correction for approval.
- Gap: recurring escalations with no source, article proposed
- Misread: tickets from customers who already read the article
- Contradiction: two articles, two answers, passages quoted
- Stale: product change referenced, affected articles listed
Why does every claim need a source?
Every claim in a proposed article carries a source, such as the product documentation, a resolved ticket, or a policy page, because a knowledge base article is the source the Digital FTE will cite to customers later. An unsourced article propagates its errors into every reply that cites it.
This also makes review fast. The approver checks that the sources support the steps rather than testing the steps from scratch, and the article records where its claims came from for the next person who maintains it.
How do you measure it?
Measure by first-line resolution rate over time, the number of open knowledge gaps, and the number of tickets per article view for the articles that were flagged. Record the baseline before starting and expect a backlog of proposals in the first weeks as the FTE surfaces the existing gaps.
The steady state is a knowledge base that grows from what customers ask, with each addition traceable to the tickets that justified it, and a support lead who sees the gaps in the morning brief rather than in the quarterly review.
What does a proposed article look like in Approvals?
A proposed article arrives in Approvals with the tickets that justified it attached: the escalations that recurred because no source covered the question, grouped, with the resolutions the team actually gave. The draft follows the knowledge base's own format, states the conditions under which each step applies, and cites the product documentation or policy page behind every claim, so the reviewer checks sources rather than testing steps from scratch.
The support lead approves, edits, or sends back with a reason, and a send-back for tone or structure becomes a rule for future proposals. Once published, the article is what the FTE cites the next time the question arrives, so the first-line resolution rate for that category is the measure of whether the article worked. Flags for stale or contradictory articles arrive the same way, with the evidence: the tickets from customers who read the article and still asked, or the two passages that disagree. The knowledge base grows from the queue rather than from a backlog, and every addition can be traced to the tickets that earned it.