lamp.
ResourcesPricingAbout

Confidence score

A confidence score is a number an AI system attaches to its answer to indicate how likely it believes the answer to be correct.

What it means

When an AI reads an invoice total or classifies an email, it can report how sure it is, often as a percentage. A business uses the score to decide routing: high scores proceed automatically, middling scores go to a person, low scores are rejected. The score is an estimate, not proof; a model can be confidently wrong. Cut-offs should be set by testing on your own past cases, not by trusting the default.

Example

A trading firm's reader extracts invoice numbers with a score. Above 95 percent it posts directly; between 70 and 95 it queues for a clerk; below 70 it asks for a clearer scan.

Records and checks

  • Record the score alongside each result and the route taken.
  • Verify by sampling auto-approved items to see whether high-score answers are actually correct.
  • The process owner reviews the cut-offs after each accuracy review.

Common mistakes

  • Treating a score of 98 percent as a guarantee, even for money-critical fields.
  • Keeping the same cut-off for every document type, though some are harder than others.
Did not find your answer?Message the lamp team on WhatsApp. Ask about setup, your records, or anything we have not covered here.Ask us on WhatsApp →