Colorado agencies report technology savings, faster services and AI safeguards
State agencies reported negotiated savings, shorter service times and artificial-intelligence tools for benefits, fraud screening and document processing as lawmakers pressed for stronger oversight.
Colorado agencies reported technology savings, faster services and artificial-intelligence tools for benefits, fraud screening and document processing during the Colorado General Assembly Joint Technology Committee’s Sept. 8 meeting.
The Office of Information Technology said it had achieved $6 million in platform savings through vendor negotiations. It also projected $1.2 million in fiscal 2027 savings for the Colorado Department of Public Safety through a staffing model and reduced overhead. The latter is a forecast, not money already saved.
OIT also reported that a change to the My Colorado sign-in process reduced access time for summer Electronic Benefits Transfer services from slightly more than two minutes to 25 seconds, or about 70%. The office said Colorado had launched a mobile driver’s license usable with the Transportation Security Administration and in federal buildings.
Benefits and fraud tools
The Department of Health Care Policy and Financing said its Health First Colorado chatbot had handled about 10,000 questions, with reported satisfaction of roughly 72% and accuracy of 89%. It answered 62% of questions without referring users elsewhere, the department said.
The chatbot draws from more than 70 approved resources rather than the open internet. It does not determine eligibility, approve or deny coverage or services, or make clinical decisions. Unresolved questions go to human staff, and users are warned not to enter protected or personal information.
HCPF said a separate provider call-center tool handles about 2,700 interactions per month using predefined conversational flows and existing Medicaid data. Officials said it does not use generative artificial intelligence or large language models and cannot make decisions about eligibility, claims, payments or provider enrollment.
The department also reported that its fraud, waste and abuse analytics platform generated 82 leads and about $200,000 in savings over six weeks after four providers were placed on prepayment review. HCPF said the platform uses human-developed statistical models rather than generative artificial intelligence. Analysts, investigators and case managers review leads before cases are opened or audits begin.
HCPF said its document-recognition system processed more than 200,000 documents in the previous 12 months with accuracy above 99.8%. Workers can correct extracted fields, and low-confidence handwritten material is routed for human review.
The Department of Labor and Employment said its Family and Medical Leave Insurance fraud-screening tool reduced investigator review time from about one hour per claim to 15 minutes for roughly 1,000 records per month. Officials said about 95% of flagged items were later judged fraudulent. That figure describes the share of flagged items confirmed as fraud, not the share of all fraud detected by the system.
Safeguards and oversight
HCPF officials described an artificial-intelligence governance committee that includes legal, privacy, technology, bias-and-ethics and program staff. They said proposed tools undergo technology, procurement, cybersecurity, privacy and budget reviews, and that an employee must review artificial-intelligence output and remain accountable for errors.
Officials also described restricted access, multifactor authentication and monthly data transfers for the fraud-detection platform, which operates in a vendor’s Amazon Web Services environment. Protected health information cannot be used without a Health Insurance Portability and Accountability Act business-associate agreement, officials said.
The Colorado Office of Information Technology’s artificial-intelligence guide requires generative-artificial-intelligence efforts and use cases, including vendor-led projects, to undergo an OIT risk assessment. The state’s strategic approach to generative artificial intelligence says the assessments consider accuracy, bias, privacy and safety using National Institute of Standards and Technology standards.
Some projects remain pilots or evaluations. The Colorado Bureau of Investigation’s proposed CaseGuard redaction tool had not been deployed and was estimated to save about 20 staff hours per month while processing roughly 75 documents.
The committee chair said the next quarterly oversight update would come in October, before the report is legally due under Senate Bill 185. She said future reports would include project numbers and costs, broken out by contractor versus employee, as well as information about ongoing Service Hub projects. The Colorado Benefits Management System was also expected to be included after lawmakers requested more information.
The chair said the committee does not have statutory authority over OIT personnel matters. Members nevertheless raised questions about staffing and transparency as the state expands technology projects and formalizes oversight.