Scaling Support With a Knowledge Base That Deflects Tickets
Key takeaway
Write articles from your actual ticket history, in the words customers use rather than your internal terminology, and surface them at the point of confusion inside the product. A knowledge base nobody can find deflects nothing, no matter how comprehensive it is.
Support volume grows with customers, and hiring proportionally is the expensive answer. Self-service content is the cheap one — but only if it's built from evidence rather than from what the team assumes people need explaining.
Write from the ticket queue
- Categorise the last few hundred tickets. The top twenty questions usually account for most of the volume, and they're your first twenty articles.
- Use the customer's vocabulary in the title. They search for "can't log in", not "authentication troubleshooting".
- Answer in the first paragraph, then explain. Support articles are skimmed, not read.
- One question per article — combined articles are harder to find and harder to maintain.
Findability is the whole battle
- Search must tolerate typos and synonyms, and rank by usefulness rather than recency.
- Publish articles publicly and indexably. A large share of help traffic arrives from a search engine, not from your help centre's front page.
- Surface relevant articles contextually in the product — on the screen where the confusion happens, not behind a global help link.
- Suggest articles in the contact form as the customer types their subject. This is the highest-deflection placement available.
Keep it current
- Assign every article an owner and a review date. Wrong help content is worse than none, because it burns trust and generates a ticket anyway.
- Update articles as part of releasing a feature change, not afterwards.
- Show helpfulness voting and read the comments — customers will tell you precisely where the article stops being useful.
- Retire articles for removed features rather than leaving them to confuse searchers.
Measure honestly
Deflection is hard to measure directly, so triangulate: tickets per active customer over time, contact rate for topics you've covered versus those you haven't, and the proportion of help searches ending without a ticket. Article views alone measure nothing — a heavily-read article beside a heavily-contacted topic usually means the article is failing to answer the question.