What is confidence score?
A confidence score is a number an AI system attaches to one of its outputs to show how sure it is, for example how likely it is that it has recognised a word or an intent correctly.
Why it matters
An AI system that treats every guess as certain will act on its mistakes. Confidence scores let a team set rules for uncertainty. The system proceeds when the score is high, asks the customer to confirm when it is middling and passes to a person when it is low. This keeps automation for the cases the system handles well and sends the rest elsewhere.
How it is measured
The score is usually a value between 0 and 1 produced alongside the output. A team sets thresholds by testing. They run labelled examples through the system and see how often outputs at each score were actually right. A score is only useful if it is calibrated, meaning that outputs scored at 0.9 turn out to be correct about nine times in ten.
Example
A caller says a sixteen-digit meter number over a noisy line. Speech recognition returns the digits with a confidence of 0.62, below the utility’s threshold of 0.85. The voice agent reads the number back and asks the caller to confirm it. One digit was wrong. The caller corrects it, and the agent continues with the right account.
