How accurate are the AI models? What about false positives?

Baseline models — deployed at go-live — are trained on broad datasets from similar operational environments. Their accuracy varies by use case and environment, but typically starts in the range of 60–75%.

Through the continuous improvement cycle, models are retrained on data from your specific site — your lighting, your equipment, your workflows. By cycle 3 or 4 of real-world retraining, accuracy commonly reaches 85–93%+.

False positives decrease significantly with each cycle as the model learns to distinguish genuine anomalies from normal variation in your specific environment.

One client’s visual inspection model reached 93%+ accuracy by cycle 4 and has operated reliably for 5+ years since.