CBMS uses Hyperscience OCR (99.8% accuracy reported) and pilots a policy Q&A assistant for staff

Nina Mack, Chief Product Officer for COBES (the Colorado Benefits Eligibility and Enrollment Systems that include CBMS and PEAK), described two state‑facing AI uses intended to speed case processing and reduce manual entry errors.

Mack said the Hyperscience ICR/OCR solution extracts data from scanned documents and converts handwriting to text; the department reported processing more than 200,000 documents in the prior 12 months and stated the process is over 99.8% accurate when it recognizes text. Extracted fields are available on CBMS screens for worker review and any changes can be reverted by a human. Mack said the system flags low‑confidence handwriting for human review rather than attempting an uncertain extraction.

The second use case, PANDA (Policy Area Natural Dialogue Assistant), is a pilot internal chatbot that answers staff questions in plain language and returns citations to documents uploaded and approved by program subject matter experts. Access is currently limited to a small group of state program staff so they can validate answers before wider release. Mack said the pilot remains limited pending further testing and as CBMS replacement planning continues.