Michigan MiDAS Unemployment Fraud Algorithm

Michigan’s Unemployment Insurance Agency launched MiDAS, the Michigan Integrated Data Automated System, in 2013 as a claims-processing and fraud-detection system. In the 2013–2015 period, the system used data matching and claimant responses to generate fraud allegations, and many findings were made through automated processes without an employee reviewing each file first. Claimants could be required to repay benefits, interest and substantial penalties; the agency could intercept tax refunds or garnish wages. A fraud decision therefore had consequences well beyond a software score. A recurring issue was the difference between time periods in employer wage reports and claimant benefit certifications. Employers reported wages quarterly; claimants certified more frequently. The Auditor General described the delayed matching process, and the state later reviewed computer-only determinations. The settlement notice in Bauserman states that the Auditor General found allegations based on auto-adjudication wrong more than 90% of the time. That figure applies to the reviewed computer-only fraud findings, not every MiDAS decision, every type of unemployment overpayment or all later state cases. Other sources cite 93% based on a particular sample; use the denominator and review period when quoting it. Claimants alleged that notices did not give meaningful reasons and that people who failed to access or answer online questionnaires could be found liable without a fair opportunity to respond. In Bauserman v. UIA, Michigan’s Supreme Court in 2022 held that plaintiffs had alleged a cognizable state constitutional due-process claim for which damages could be sought, and remanded the case. It did not rule that every MiDAS determination was unlawful. A civil-rights class settlement was later approved for eligible claimants. MiDAS was not simply “a buggy program.” The failure combined automation, poor matching and notice, performance incentives, weak human review, collection before an effective hearing and inadequate process design. Safer systems distinguish inconsistency from intent, explain the evidence, allow accessible correction and appeal, and pause collection while a timely dispute is reviewed.