900 articles, 28% self-service: pruning a fintech knowledge base
Northbridge Pay is a payments platform with 900 external knowledge base articles across five product lines, a customer portal, and two technical writers. Self-service resolution is 28%, and 46% of tickets are for issues that already have an article. Search analytics show the top 20 queries return an average of six results, three of which are outdated or duplicate. A fee-policy change in March required edits to 12 articles; by June, support agents were pasting links to four old versions. Compliance review takes 10 business days, and writers are measured on articles published per month. The head of support wants to commission 150 new articles to close coverage gaps. The portal search is the only self-service entry point. Agents can see internal articles but customers cannot.
1. Should Northbridge write 150 new articles? Diagnose the self-service problem and prioritize the next 30 days.
Read the worked answer
No. Publishing 150 new articles would add to a search problem that is already producing six results for top queries, three of them outdated or duplicate. The stronger signals are 46% of tickets having an existing article and agents pasting old versions after a policy change. That is a findability, trust, and lifecycle problem, not a coverage-only problem. Writers measured on articles published per month are incentivized to add volume rather than remove duplicates.
Next 30 days:
- Join the top 20 ticket reasons to the top 20 portal searches and the articles agents link most.
- Consolidate duplicates and retire or redirect stale articles, starting with fee-policy content.
- Rewrite the top 20 queries to one canonical article each, with a clear title, summary, and next step.
- Add a feedback control and review search terms with no clicks.
- Batch compliance review with pre-approved policy templates rather than one-off 10-day reviews.
- Change writer goals to article health, search success, and ticket deflection, not publish count.
New articles only for validated gaps where no canonical article exists.
2. Design a knowledge governance model that prevents another stale fee-policy link problem.
Read the worked answer
Governance should make the canonical article the only customer-facing source. Assign an owner per product line and a backup. Every article gets a review date, a status (current, needs review, retired), and a change log. Policy changes trigger a named task list, not a general request; legal/compliance keeps pre-approved language blocks that writers can assemble without a new 10-day review for every edit. Retired articles redirect to the replacement and are removed from agent macros. Agents link only to the canonical customer article; internal notes stay internal. A monthly sample audit checks top 50 articles for stale links, wrong policy, and duplicate titles. Writer and product incentives should reward accuracy and deflection, not volume. This prevents the March fee-policy failure: one owner, one canonical article, a redirect from the old version, and a compliance-approved block.
3. How would you measure deflection credibly without overclaiming?
Read the worked answer
Measure deflection with a chain, not a pageview. Define self-service resolution as a customer who views a canonical article or completes a portal flow and does not open a ticket for the same issue within seven days. Track portal sessions, search success (click and no immediate re-search), article feedback, and tickets tagged with a reason of self-service attempted. Compare tickets per 1,000 active customers for the top 20 issues before and after the rewrite. A staged rollout by product line gives a comparison group; a simple before/after does not prove causation because seasonality and product releases move ticket volume. Report the limits: a customer may solve the issue without the portal, and a ticket may be created for an unrelated reason. Use the combination of search success, top-query ticket volume, and customer feedback as evidence, not pageviews alone.