The knowledge base deflection rate nobody talks about
Most teams measure whether players use the KB. Few measure whether the KB answers the question they actually had.
Deflection rate is the most cited KB metric and the least useful one in isolation.
"We have a 30% deflection rate" means 30% of players who visit your help centre do not submit a ticket. It does not mean they found their answer. It does not mean they were satisfied. It means they left without a ticket — and you are counting that as success.
What actually happened when a player left without a ticket
Three outcomes are possible when a player leaves your help centre without contacting support: they found their answer and resolved their issue, they gave up and accepted the loss, or they found the answer elsewhere (Discord, Reddit, a friend) and you have no visibility into what they needed.
The third outcome is the most common and the most dangerous. A player who resolves their issue through community channels is a player your KB failed. They did not deflect — they went around you. Your deflection metric recorded a success.
How to measure actual resolution, not deflection
The proxy metric that correlates with actual KB resolution is return rate. A player who found their answer does not come back to the same article within 48 hours. A player who did not find their answer returns, or abandons and raises a ticket within 48 hours. Track article exit → ticket submission within a 48-hour window by article. The articles with the highest rates are failing their readers.
The second metric is search exit rate. If players search a term and leave without clicking any result, your KB does not have an article that matches what they searched. Build the article. The search data is your roadmap.
The content quality gap
Most KBs fail because they are written for the team that knows the product, not the player who doesn't. The article assumes knowledge the player doesn't have. The steps are written in the order they're easiest for the author, not the order they're easiest for someone unfamiliar with the UI.
The clearest signal of this problem: an agent answers the same question in a support reply that is already covered by a KB article. That means the article exists but isn't working. The resolution exists but isn't accessible. Fix the article, not the deflection metric.
Ready to try CustomerOps Cloud?
Start a free 14-day trial and see how it works with your actual tickets.
No credit card required · Cancel anytime · Full access from day one