What is the confidence score based on?

How well the retrieved content matches the question. Weak or thin retrieval lowers the score, which triggers a handoff instead of a risky answer.

Can I test the handoff behaviour before going live?

Yes. The dashboard playground lets you simulate conversations and see when the assistant answers versus when it escalates.

Does the customer have to repeat themselves on handoff?

No. The full conversation transfers with the escalation, so your team picks up with all the context already in front of them.

Reliability

Confidence handoff: escalate to humans at the right moment

CertifChat scores its confidence on every answer. When that score is low, it says it is not sure and hands the conversation to your team with full context, so customers get a straight answer, not a confident wrong one.

Join the waitlist See pricing

What you get

Answer-confidence score computed on every reply
Low-confidence questions trigger a clear "not sure" instead of a guess
Handoff carries the full conversation so customers don't repeat themselves
Escalation rate and knowledge gaps tracked in analytics
Tunable behaviour you can rehearse in the test playground

How it works

Step 1

Score every answer

Based on retrieval quality and how well the content matches the question, each reply gets a confidence score before it is sent.

Step 2

Hand off when unsure

Below the threshold, the assistant stops guessing. It tells the visitor it is not certain and routes the conversation to your team with the full history attached.

Step 3

Learn from escalations

Every handoff is logged as a knowledge gap, giving you a prioritized list of content to add so fewer questions need a human next time.

Related features

Explore more of what your assistant can do.

Frequently asked questions

Put your knowledge to work today

Connect your sources, drop one snippet on your site, and start deflecting tickets. We are opening CertifChat to a first group soon.

Join the waitlist See pricing