Step 1 (days 1–4): connect, categorise, and record the baseline
Connect the helpdesk and the knowledge base to the Customer Support Digital FTE with scoped credentials: read tickets and articles, propose articles, never publish. Record the baseline per category: tickets per week, first-line resolution rate, escalations for no source found, and, where the helpdesk tracks it, tickets opened by customers who had already viewed an article. Those last two are the cleanup's targets.
Set the approved sources for articles: product documentation and policy pages, so every proposed article can cite where its claims come from.
Step 2 (days 5–10): flag what the tickets prove is wrong
The FTE reads a month of tickets against the knowledge base and raises detections with evidence: articles that customers read and still opened a ticket about, which are misread or wrong; pairs of articles that give different answers to the same question, with both passages quoted; and articles referencing behaviour a product change made stale. Each detection lists the tickets that prove it.
Triage the detections with the support lead: fix now, fix later, or dismiss with a reason. Dismissals teach the FTE what the team does not consider a problem.
- Misread: viewed the article, still opened a ticket
- Contradiction: two articles, two answers, passages quoted
- Stale: references a changed product behaviour
- Gap: recurring escalations with no source
Step 3 (days 8–20): approve corrections and proposed articles
For each fix-now detection, the FTE drafts the correction or the new article from how the team actually resolved those tickets, with every claim citing the documentation or policy page behind it, and submits it to Approvals with the tickets attached. Review by checking the sources support the steps and the tone matches the base; send back with a reason where not. Approved articles publish through the knowledge base's own workflow.
Prioritise by ticket volume: the article behind the most tickets is fixed first.
Step 4 (days 15–25): let the new articles work, and watch the categories
As corrected and new articles go live, the FTE cites them in first-line replies, and the categories they cover should show first-line resolution rising within days. Escalations for no source found in those categories should fall. Read the weekly detection list; it shortens as the backlog clears, which is the visible sign the cleanup is working.
Where a category does not move, look at the tickets: the article may be right but unfindable, which is a structure problem the FTE can also flag.
Step 5 (days 26–30): measure against the baseline
At day 30, compare per category: first-line resolution rate, escalations for no source, and tickets per article view where tracked, against the baseline. Count the articles corrected and added, each traceable to the tickets that justified it. The categories that moved are the cleanup's result; the ones that did not are next month's detections.
Set the steady-state rhythm: the FTE proposes articles for recurring gaps continuously, and the support lead approves them at the weekly read.
What does done look like?
Done is a knowledge base that grew and was corrected from the tickets rather than from a backlog, with every addition sourced and approved by a person, first-line resolution visibly higher in the categories that were fixed, a short weekly detection list, and a support lead who spent an hour a week approving articles instead of a quarter rewriting a wiki.
The base is now a reliable source for the fact gate, which is what makes the FTE's replies safe to promote later.
What should you refuse to do?
Refuse a general rewrite; fix what the tickets prove and leave the rest. Refuse to let articles publish unattended, because the base is a source and a wrong article propagates into every reply that cites it. Refuse to accept an article without sources. And refuse to measure by article count; measure by first-line resolution in the categories the articles cover.
The refusals keep the cleanup evidence-led, which is the only kind that lasts.