lamp.
ResourcesPricingAbout

Uncertainty handling

Uncertainty handling is the set of rules for what an AI agent does when it cannot find or trust an answer: ask, flag it or pass it to a person.

What it means

Guessing is the most dangerous behaviour in an agent, because a guessed amount looks as real as a correct one. Good handling makes "I am not sure" an allowed result. The agent leaves the field blank, states what is missing, asks the sender or routes the item to a reviewer. Rules should name fields never to be guessed, such as amounts, bank details and party identity.

Example

An agent reading a delivery challan for a plastics trader cannot make out whether the quantity is 180 or 160 kg. It marks the field unclear, attaches the image and sends it to the stores clerk.

Records and checks

  • Record each unsure case, the reason and who resolved it.
  • Verify that unsure fields are blank or flagged and not filled with plausible values.
  • The process owner reviews the rate of unsure cases, which shows where inputs or instructions are weak.

Common mistakes

  • Rewarding the agent for always giving an answer, which pushes it to guess.
  • Letting unsure cases pile up in a queue nobody owns.
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 →