1,100 Articles, 12% Deflection: Fixing Supply Before Search
An Employee Center carries 1,100 HR knowledge articles. Self-service deflection is 12% and 34% of HR searches return zero results; the top search term is "parental leave." Five agents resolve about 200 cases a month and are told to write articles, but the HR knowledge base's Can contribute criteria lists only the knowledge manager, so the drafts they create never leave Draft state. Roughly 400 articles have Valid to dates in the past. No HR service has any knowledge attached on its record, and similar-case/agent-assist suggestions are switched off in the agent workspace.
1. Rank the four symptoms by their impact on deflection and justify the ranking.
Read the worked answer
Deflection is supply-driven before it is search-driven, so rank by how much each symptom removes usable answers:
- 400 expired articles plus the contribution bottleneck. Together these are the supply. Expired articles are invisible; a KB only the knowledge manager can contribute to means the people who actually know the answers cannot publish them. With five agents producing 200 cases a month of raw material sitting in Draft, the corpus is actively decaying.
- No knowledge attached to HR services. This breaks deflection at the intake moment, when the agent is creating the case and could suggest the answer before work starts. It also starves the self-service suggestions shown alongside the service.
- 34% zero-result searches. Partly downstream of (1) - the article that would answer "parental leave" may be unpublished - and partly vocabulary mismatch between employee phrasing and article titles.
- Agent-assist/similar-case switched off. Valuable, but it accelerates agents who already have articles to find; it cannot compensate for missing content.
The judgement being tested: do not start with search tuning. Tuning relevance on a thin or expired corpus produces a measurable improvement in zero-result rate and almost none in deflection.
2. How do you fix the contribution bottleneck without opening the knowledge base to everyone? Explain the control that replaces the current restriction.
Read the worked answer
Replace a person-based restriction with a group + review control:
- Can contribute = the HR agent group, not an individual. That is the whole unblocking move.
- Can read stays as-is unless the content must reach employees; keep sensitive KBs separate with tighter read criteria rather than withholding contribute rights across the board.
- Ownership and a publish step - an article template per HR topic, a named reviewer per topic area, and a published target (e.g., five business days from draft to published or rejected). Without a publish step, freeing contribution just moves the backlog from Draft to "unreviewed."
- Create-article-from-case with pre-filled short description and topic, so contribution costs an agent two minutes, not twenty.
- Feedback loop - rejected drafts get a reason; stale articles get an owner and a review date.
Measure: draft age (median and 90th percentile), publish rate per agent, article reuse, and - the real outcome - inquiries deflected per topic. Do not measure "number of articles" as a success metric; 1,100 existing articles are precisely the evidence that count is not the problem.
3. How would you make the right article appear at the moment a case is being created for a specific HR service?
Read the worked answer
Attach the relevant knowledge or knowledge bases to the HR service's knowledge related list. That is what makes suggestions contextual: an agent opening a parental-leave case sees parental-leave content, not the whole 1,100-article corpus. The service is the natural anchor because it already encodes what the case is about.
Then verify the two access layers separately, because they fail differently:
- Agent side - the KB's Can read must include the agent group, and the article must be Published and within its valid dates. Test by opening a case as an agent for that service and confirming suggestions appear.
- Employee side - the KB's read criteria and any HR-criteria audience targeting must match real employees. Test by impersonating an employee with a complete HR profile and one with a sparse profile; if the article shows for the first and not the second, you have found a targeting problem, not a knowledge problem.
Keep the per-service attachment list short - five to ten genuinely relevant articles. Attaching an entire knowledge base to every service reproduces the needle-in-haystack problem the attachment was meant to solve.