From the State Government edition of September 25, 2026
Michigan Digital States interview reports MDHHS document-AI hours saved and coding-assistant gains
Michigan Department of Technology, Management and Budget (DTMB) / Acting CIO Eric Swanson · State shared services / health-human services benefits operations / enterprise IT · Michigan, United States
- Publisher
- Government Technology — Digital States 2026 ‘A’ State Interview: Eric Swanson, Michigan
- Original publication
- 2026-09-23
- Source retrieved
- 2026-09-26
- Event date
- 2026-09-23
What happened
Why now: 2026-09-23 Digital States A-state interview (day before the archived Digital States 2026 overview/Utah Gemini piece) gives Michigan Acting CIO Eric Swanson’s measured GenAI results. MDHHS AI document processing for food-assistance renewals claims up to 97% automation on digital forms and up to 73% on handwritten forms, cutting an estimated 266 staff hours per week while eligibility decisions stay under program rules and staff review. DTMB’s AI coding assistant claims include updating .NET dependencies across 400+ files in two hours with zero errors, embedding WCAG 2.1 AA rules, rewriting classic ASP to .NET 10/Angular 21, and automating 165 of nearly 800 critical/high vulnerability remediations across 80 components. Separately, MSP Firearms Records Unit OCR claims ~241,800 records processed, ~42% auto-approval, ~101,500 manual reviews eliminated—public-safety adjacent; transfer only with stream limits.
Why it matters
Concrete state HHS benefits-operations and enterprise IT productivity evidence from an A-grade CIO interview in the same Digital States wave as yesterday’s edition—directly transferable to other states’ SNAP/Medicaid document intake and SDLC copilots, with explicit human-in-the-loop eligibility language.
Evidence and measured results
GovTech Digital States Q&A quoting Swanson/DTMB program figures; operator-reported, not an independent statistical evaluation. MiCARA SNAP case-review tool described as supporting specialists. MSP firearms figures included in same answer but tagged as public-safety adjacent.
Limitations and uncertainty
Automation rates and hour-saved estimates are state-reported via journalism. MSP firearms OCR is public-safety/policing-adjacent—do not treat as core benefits evidence without stated limits. No baseline methodology or error-rate audit published here.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-25; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- State HHS and IT leaders need credible, human-supervised GenAI productivity metrics for benefits document intake and modernization without promising unverified FTE cuts.
- Stakeholders
- State CIO / DTMB
- MDHHS eligibility leadership
- Enterprise architecture / AppDev
- Privacy and security officers
- Budget/legislative staff
- Discovery questions
- How is the 266-hour/week estimate measured and audited?
- What is the false-accept rate on handwritten renewals?
- Who owns override when IDP extraction conflicts with case rules?
- Are coding-assistant vulnerability fixes independently retested?
- Value hypothesis
- A bounded IDP or SDLC-copilot measurement engagement can reproduce Michigan-style hours-saved evidence under explicit human-review gates.
- Potential engagement
- 6–8 week benefits-document IDP pilot design with baseline time-motion and error sampling.
- Unsupported claims
- Guaranteed statewide FTE reduction
- MSP firearms auto-approval as benefits proof
- Zero eligibility error from 97% automation
Pre-sales engineering
Role takeaway
- Fit
- Strong fit for renewal-form extraction and constrained code remediation with human gates; unfit as sole eligibility decision engine.
- Architecture and integration
- IDP over digital/handwritten forms into MDHHS workflows; AI coding assistant on .NET/Angular repos; MiLogin/zero-trust identity context.
- Prerequisites
- Labeled forms corpus
- Eligibility specialist review queue
- Secure code-assist policy and secret scanning
- PII minimization in document stores
- Constraints
- Michigan program rules
- Federal SNAP accuracy pressures
- Operator-reported metrics only
- Public-safety MSP metrics out of benefits scope
- Security
- Document PII controls for IDP; require human disposition logs for AI-proposed vulnerability patches.
- Proof of value
- Replicate hours-saved on sample caseload
- Handwriting error/false-accept audit
- Accessibility regression on remediated apps
- Independent retest of AI-applied security fixes
Delivery
Role takeaway
- Work
- Forms taxonomy, override UX, measurement dashboards, and secure SDLC playbooks for AI assistants.
- Dependencies
- MDHHS program policy owners
- DTMB platform/AppDev
- Identity/cyber teams
- Training for specialists
- Ownership
- DTMB enterprise AI with MDHHS operational ownership of eligibility outcomes.
- Skills and adoption
- Specialists need override/QA skills; developers need AI-assisted code review discipline.
- Governance checkpoints
- Pre-scale accuracy gate on handwritten forms
- Monthly false-accept review
- Security disposition audit
- Accessibility WCAG checks
- Acceptance criteria
- Published human-review rate
- Hours-saved methodology documented
- No unsupervised eligibility auto-certification
- Vuln-fix retest evidence
- Risks
- Silent extraction errors altering benefits
- Scope creep into decisioning
- Misuse of MSP metrics in civilian procurements
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Intelligent document processing over digital and handwritten forms feeding MDHHS renewal workflows; human eligibility certification retained; enterprise AI coding assistant in .NET/Angular modernization; MiLogin/zero-trust identity context referenced for cyber strategy.
Governance
Who approves, reviews and stays accountable for outcomes?
DTMB-led enterprise AI strategy; SNAP payment-accuracy pressure cited as federal driver; human-in-the-loop validation emphasized for MiCARA and MDHHS IDP; accessibility remediation tied to WCAG 2.1 AA.
Security and privacy
What data, permissions and controls need testing?
Vulnerability remediation automation with documented dispositions; zero-trust and MiLogin realignment noted; benefits-form IDP implies sensitive PII—controls not detailed in interview.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
WCAG 2.1 AA remediation and AI-assisted accessibility rules are cited alongside IDP that reduces manual entry so specialists spend more time on judgment work—not a job-cut claim.
Procurement
What should contracts, pricing and exit terms secure?
Interview does not name IDP or coding-assistant vendors; discovery should identify platforms, data-processing terms, and federal program compliance constraints.
Operating model
Which teams own the service once it runs?
Enterprise AI strategy under DTMB with agency operational owners; human-in-the-loop eligibility and case review remain explicit; Digital States interview used as external reporting channel for metrics.
Publication history
- 2026-09-25State Government · Issue 204 resources
Stable resource ID: michigan-digital-states-2026-swanson-mdhhs-idp-20260923