Colorado says AI sped Universal Preschool income checks and cut costs

Colorado early-childhood officials told lawmakers Monday that a new Snowflake Cortex workflow is helping staff process Universal Preschool income-verification documents faster and far more cheaply than a vendor-built system, while leaving final eligibility decisions to people.

At a Joint Technology Committee hearing, Colorado Department of Early Childhood officials said the program had handled about 10,000 applications this year. They said specialist throughput rose from roughly 200 to 300 applications a year to about 175 a week. Officials also said the in-house setup cost about $5,600, compared with an estimated $100,000 for a vendor system, and that monthly operating costs were in the single digits.

The department described the tool as a backend document-extraction system, not a public-facing chatbot or automated decision-maker. Families upload income documents through BridgeCare, and CDEC retrieves them by API into Snowflake, where AI extracts fields for staff review. Officials said the system is used only for this Universal Preschool verification workflow, not for broader eligibility decisions across the department.

Officials said the AI does not make final eligibility decisions. If an expected data field is missing, the case is routed to a person rather than auto-approved or auto-denied.

Lawmakers also asked about safeguards against malicious or malformed uploads. CDEC said the workflow is not externally facing, users cannot enter prompts, and backend prompting is limited to developers with strict permissions. The hearing record did not show a formal independent security review.