The Exemption Is a Query: Who Actually Decides Minnesota’s Medicaid Work Requirement

On January 1, 2027, Minnesota’s Medical Assistance program begins requiring some adults to prove they work. The day before, the last federal recovery dollars from the American Rescue Plan Act expire, taking with them money that has carried a good share of city and county public health, violence prevention and emergency planning work since 2021.¹ The overlap has been covered as a budget story, and it is one. It is also a story about who makes decisions. Most eligibility determinations under the new requirement will not be made by a person reading a file. They will be made by a data match, and the people who answer for those determinations at the county counter will not have seen the logic that produced them.

What the rule requires

Under H.R. 1 and the interim final rule the Centers for Medicare & Medicaid Services issued on June 1, 2026, expansion adults must show 80 hours a month of work, community service, participation in a work program or half-time enrollment in school or career training, or monthly income of at least $580.² States must look back at least one and up to three months at application, and must verify at least one compliant month in each six-month review period. Renewals for this group move from once a year to twice.³ Before a state may ask a person for paperwork, it must first check the data it already holds, such as payroll data and Medicaid claims and encounter data. If it cannot verify compliance, it sends a notice of noncompliance and the person has 30 days to show compliance or an exemption before coverage is denied or ended. Self-attestation is permitted through 2027; from January 2028, documentation is generally required where it is reasonably available.³

The first decision is made by a query

The Minnesota Department of Human Services told MPR News in August that about 243,000 Minnesotans on Medical Assistance will have to meet the requirement.⁴ The department’s 2026 supplemental budget estimated that about 128,000 people will be subject to regular reporting.⁵ The two figures come from different documents and may not rest on identical assumptions, but the most direct reading of the gap is that more than 100,000 people are expected to be cleared without ever being asked, because a data source will show them working, exempt or otherwise compliant. That is the program working as designed, and it is far better than making a quarter of a million people file paperwork twice a year. It also means the largest single determination in the program, who must report and who need not, is a system output. Nobody in a county office signs it.

The state is building toward exactly that. A March 2026 DHS systems modernization briefing to the Legislature schedules “METS functionality improvements tied to HR-1” between the third quarter of 2026 and the first quarter of 2027, lists intelligent document processing, workflow automation and call-center virtual assistants as under analysis, and describes a competitive “bake-off” to select the eligibility rules engine.⁶ The same briefing reports that automated ex parte renewals rose from 17 percent of renewals to above 76 percent during the Medicaid unwinding, saving an estimated 115,725 worker hours. It also states that “demand for systems work exceeds available resources,” and that the department’s IT budget is 1.3 percent of operating expenses against a peer average of 4.1 percent.⁶ None of this is a criticism of DHS staff. It describes the conditions under which the automation will be built: quickly, on a thin budget, against a fixed federal date, with federal guidance and litigation still moving.

The vendor market has arrived to meet those conditions. The Center on Budget and Policy Priorities reviewed well over a dozen products aimed at work requirement verification, from Equifax’s The Work Number at roughly $15 per income report to per-verification and per-enrollee tools, several of them with AI document classification or chatbots built in.⁷ Its caution about single-vendor arrangements applies to every county that will live with the result: the risk of “elevated costs, diminished control over implementation outcomes, and challenges in enforcing vendor accountability.”⁷

Where the data runs thin

Automated verification works best for people whose lives already sit in clean records: one W-2 employer, a payroll provider that reports to the database, a condition coded the way the rule expects. It works worst for the kinds of work Greater Minnesota has a great deal of. CBPP notes that the traditional income databases do not cover all W-2 employees and do not cover 1099 workers.⁷ Farm labor, seasonal work, small employers paying in cash, gig work and self-employment are where a match fails and a notice goes out to someone who met the requirement. Medical frailty, the broadest exemption, is the category data sees least well. Danny Mintz of Code for America told Nextgov that there is “not really necessarily a good data source that covers those cases at this point.”⁸ The rule directs states to maintain “auditable, regularly updated” ICD-10 code lists to identify people who may be medically frail.⁹ A code list is a model. Someone decides what goes on it, and people whose disabling conditions are undiagnosed, miscoded or treated outside the claims system are the ones it will miss.

Arkansas has already run this experiment. When the state imposed work reporting in 2018, about 18,000 of roughly 100,000 targeted adults lost coverage within months. A Harvard team writing in the New England Journal of Medicine found that more than 95 percent of the targeted group were already working or should have been exempt, and that about a third had never heard of the requirement. Employment did not rise.¹⁰ People lost coverage because the reporting system failed them, not because they failed the test.

The county holds the remainder

Every case the data cannot clear lands with a person: the noncompliance notice, the 30-day window, the phone call, and the visit to the county or Tribal office that DHS now tells enrollees to make.¹¹ Minnesota’s last eligibility surge, the Medicaid unwinding, required $36 million in state funding to help counties and Tribal Nations handle the additional work.⁶ The work requirement returns that load on a six-month cycle, in the same year that ARPA-funded positions end. When eligible people lose coverage anyway, the cost resurfaces as uncompensated care at safety-net hospitals, which is the pressure the Next City and MinnPost reporting documents at Hennepin County Medical Center.¹

That leaves county human services directors in a difficult position. Their staff will explain, defend and sometimes reverse decisions produced by a rules engine and a set of data sources the county did not choose, cannot inspect, and may not be told about when they change. When a match is wrong, the eligibility worker at the counter becomes the public face of an error made somewhere else, and the county absorbs the appeal, the reinstatement and the complaint.

Five questions to put to the state before January

Counties do not control METS or the rules engine. They do control what they ask for before the start date and what they insist on measuring after it. Five questions are worth putting to DHS in writing now, through MACSSA or directly.

1. Coverage of the data. Which data sources clear a person automatically, and which groups do those sources not reach? The answer should address self-employed, farm, seasonal, gig and cash-paid workers by name.
2. Review before notice. Before a noncompliance notice is issued, does a person review the case, or is the notice generated directly from a failed match?
3. Error reporting. Will counties receive monthly figures on terminations for noncompliance, and on how many of those were reversed on appeal or reinstated within 90 days? Reversal rates are the cheapest error measure available, and the state already holds them.
4. Vendor and AI visibility. Where a vendor product or AI feature touches the determination, whether document classification, a chatbot answering eligibility questions or a screen for likely frailty, what can the county see of its rules, and who audits it?
5. Change control. When CMS guidance changes or a court rules on the frailty definition, how quickly can the logic change, and how will counties be told before their staff are answering for it?

None of these questions delays implementation. Each one makes an error visible sooner, which is the only dependable way to keep a verification system from functioning as an eligibility cut that no one voted for. The federal rule fixed the start date. What counties ask for in the next twelve weeks will decide how much of the system they can see once it is running.