Deflection has always been a proxy metric — you can't directly observe a ticket that was never created. But there are better and worse ways to estimate it.
The kb_use table captures article interactions including the "This resolved my issue" flag on the portal. It's the basis for most OOTB deflection reporting, but it's self-reported and most users who find what they need don't click anything — so it undercounts.
Combining a few sources gives you something more defensible:
- Search sessions without ticket creation — queries that end without an incident or case submitted within a defined window (24–48 hours is common) are a reasonable deflection proxy
- Portal session analysis — KB article views followed by session end without ticket submission
- Ticket volume trending by topic — if knowledge coverage improves in an area and ticket volume drops, that's a meaningful signal even if not a precise count
Some estimation is unavoidable. What matters more than finding a perfect metric is agreeing on a consistent methodology with your stakeholders before you start measuring — otherwise the number gets challenged every time it's presented.
