Let Valen route the clear cases
Valen returns probabilities over your candidate intents. Use them to route qualifying tickets automatically and send the rest for human review.
See how much work it can take on
In a retrospective replay of BANKING77 predictions, Valen 0.8B accepted 147 of 1,276 test questions using a rule selected before the test. One accepted answer was wrong: 11.5% automated coverage with 0.68% accepted error.
This is an English benchmark with 77 candidate intents. Explore the released-model evidence to compare models and validation targets, inspect accepted counts and errors, and check the study's scope.
Choose the scope using your own data
Calibration uses labeled examples from your workflow to check and adjust the reported probabilities. The tools report automated coverage, errors and review volume, helping you decide whether to integrate Valen.
Fitting calibration parameters requires no model retraining. Save the parameters and routing policy for the same model and runtime configuration, and decide whether to adopt them using your independent test results.
Run the complete ticket-routing example →
The example includes the data format, calibration commands and routing code. It returns automatic or human_review for integration with your existing queues.