How does Kav handle uncertainty? What it does when it does not know
When Kav has no approved answer it says so instead of guessing. How it fails safe, where answers come from and how a person takes over.
It says so
When the assistant has no approved answer, it declines plainly instead of guessing. A wrong answer about prices, hours or eligibility costs more than an honest "I don't have that."
Fail closed, by design
Everywhere in the platform, the safe outcome is the default. A rule condition that cannot be evaluated does not match. A missing permission denies. If something fails, the assistant gives the configured fallback and the failure is raised for your team to see. It does not fall back to unrelated content.
Answers come from your approved content
The assistant answers from the knowledge you maintain. Content goes live only after the accountable person approves it, and the console shows what is waiting.
The model never writes the numbers
Amounts, dates, reference numbers and statuses are filled in from system data into fixed templates. The model chooses the template and writes the words around it, so it cannot invent a figure.
Handing over to a person
When a question is out of scope or the customer asks for a human, the conversation moves to your team with the whole thread attached, so nobody asks again.
Closing the gap
Unanswered questions show up as content gaps. Add an article, approve it, and the next customer gets the answer. This is how coverage grows without engineering work.
What to expect
A correct decline will sometimes feel like a missing feature. It is the platform behaving as designed.