How accurate is Industry Prediction, and how should I use the accuracy score?

Last updated: September 14, 2026

Every prediction comes with an accuracy score from 0 to 1 that reflects how confident we are in the code. Predictions made from a live website with full analysis score highest. Predictions made from the name and address alone, with no website found, score lower because there is less to work from.

How to use the score.

  • Treat 0.75 and above as reliable for automated decisioning. Predictions at this level are very rarely wrong at the sector level and usually right at the six-digit level.

  • Route predictions below 0.75 to review, or fall back to a coarser code. Truncating to the four-digit industry group is often correct even when the six-digit code is uncertain.

  • Use keywords and reasoning when reviewing. They show what the model saw, which makes a wrong prediction quick to spot and a right one quick to confirm.

  • Use risk_level as a separate signal. A confident prediction into a high-risk industry and an uncertain prediction into a low-risk one call for different handling.