What Medicaid Unwinding Teaches About Eligibility System Readiness

Start with the documented record

During the pandemic, enhanced federal Medicaid funding was conditioned on maintaining enrollment for most beneficiaries, subject to exceptions. The continuous-enrollment condition ended March 31, 2023. States could begin unwinding renewals in February, March, or April 2023, with terminations no earlier than April 1. CMS initially allowed 12 months to initiate renewals and up to 14 months to complete them; subsequent flexibilities, pauses, and pending cases affected completion schedules. A state's last scheduled renewal cohort did not necessarily mean every outstanding determination was finished. CMS’s May 2024 timeline analysis.

GAO’s July 2024 review provides specific findings worth carrying into acceptance testing:

• Individual decisions: CMS identified 29 states that were not conducting ex parte reviews at the individual level. CMS estimated that about 420,000 eligible people, including children, lost coverage as a result.
• Population coverage: 26 states were not prepared to conduct ex parte reviews for certain groups, including people with disabilities or over age 65.
• Submission channels: 19 states did not allow renewal forms through all required modalities.

Those categories can overlap. They should not be added together or treated as a ranking of states' technology. GAO-24-106883.

October 2026 review update: GAO now marks its oversight recommendation implemented as of January 2026, citing CMS guidance, continuing data monitoring, and eligibility oversight programs. This is a meaningful federal corrective step; it does not certify every state's system or settle every individual case. The state-level task is to retain evidence that the relevant defects have been corrected and remain covered by regression tests.

Ex parte renewal is a decision process, not an automation score

An ex parte renewal uses reliable information already available to the state without asking the beneficiary for additional information. CMS’s November 2024 guidance expressly includes automated and manual steps. A process can therefore be ex parte without being fully automated, and a high automation rate does not by itself establish that decisions are correct. CMS ex parte guidance, pages 3–4 and 10.

For engineering teams, begin with a traceable path from person to evidence to decision. Household data may inform an individual's eligibility, but a result for one member must not silently become the result for everyone. Test households whose members have different eligibility categories, income thresholds, or verification needs. Include populations that require information beyond the most common income checks.

Treat differences in renewal rates as a starting point for investigation. Before comparing states, agencies, or releases, establish the reporting period, eligible population, numerator, denominator, pending-case treatment, and meaning of “automatic.” Then investigate data coverage, matching, rules, manual steps, and operational practices. A rate alone cannot establish that APIs caused better results or that population differences had no effect.

Useful diagnostic questions include:

• Was useful evidence available but never requested or read?
• Did a source respond, but a mapping or identity match fail?
• Was the household assembled correctly for each person's eligibility category?
• Could staff complete an ex parte review with available evidence when automation could not?
• Did the process request only the additional information needed when renewal could not be completed?

The answers identify work that can be tested, owned, and prioritized.

Keep contact information usable throughout the case

CMS’s January 2023 unwinding guidance tied enhanced-funding conditions to efforts to update contact information and handle returned mail. It allowed NCOA, state-agency information, or other reliable sources; NCOA was not the sole mandatory method. The guidance also recognized that a postal address source does not provide telephone numbers or email addresses and may not contain a current address for every person. These particular funding conditions applied during specified quarters of 2023. Current operations must use the requirements and approvals applicable to the case, rather than assuming a temporary unwinding provision is permanent. CMS SHO 23-002, pages 10–15.

The durable engineering lesson is to maintain contact data as part of the case record. Record where an address or number came from, when it was confirmed, the person's communication preferences, and whether delivery succeeded. Provide a controlled way to reconcile conflicting updates. A staff correction should reach the systems that generate notices, reminders, and case status messages; otherwise a correct field on one screen may leave the next mailing unchanged.

Design for more than the portal. Test paper, telephone, in-person, and electronic handoffs as applicable, including accessible documents and assistance. Evaluate language, document format, connectivity, and accessibility as distinct service conditions. None should change the substantive eligibility standard.

A useful hypothetical test case combines an unsuccessful data match with returned mail. The workflow should retain the individual's deadline, contact attempts, outstanding evidence, and assigned owner together. Otherwise two teams may resolve separate fragments while neither completes the case. The test checks coordination; it does not presume that these were the largest or most expensive exception categories nationally.

Test surge capacity without guessing at the cause

A major renewal campaign can increase demands on data sources, document processing, staff queues, notices, and call centers. Whether a particular system fails depends on its measured workload and constraints. Investigate capacity limits in the actual environment, including batch processing windows and shared resources, before selecting a replacement design.

Use a production-informed load model. Include new applications, renewals, reported changes, appeals, and rework at the same time. Document assumptions about arrivals, processing time, source availability, worker capacity, and deadline distribution. Then test the conditions most likely to change those assumptions:

1. A dependency slows down. Does the system retry safely without flooding the source or duplicating a determination?
2. Documents arrive in a burst. Can intake acknowledge receipt and preserve the deadline while classification or staff review catches up?
3. A processing window overruns. Are worker screens and notices protected from partial or inconsistent results?
4. One workload consumes capacity. Can urgent cases continue while bulk processing is limited or rescheduled?
5. A release must be reversed. Can operators identify affected cases and resume work without losing evidence or issuing contradictory notices?

Separate queues or processing pipelines can help isolate workloads where testing demonstrates contention. Define capacity limits and priority rules explicitly, and prevent lower-priority work from waiting indefinitely. Shared databases, external services, and staff teams may remain common bottlenecks even after queues are separated. Measure the complete process before claiming the architecture has solved the problem.

Choose integrations by evidence quality and operating needs

State eligibility teams may draw on wage records, SSA information, other benefit-program information, and other authorized sources. CMS’s ex parte guidance discusses useful sources and verification plans, including quarterly wage data. Source access, legal permissions, reliability, and the eligibility factor being tested matter alongside transport technology. CMS guidance, pages 8–10.

An API can shorten delivery latency when the upstream source exposes timely information. It cannot make quarterly data current to yesterday or establish access rights that the agency does not have. A scheduled file can remain appropriate when its content, delivery schedule, and recovery process meet the business need. For federal and state integrations alike, specify:

• Evidence: source, covered population, reporting period, known gaps, and matching confidence.
• Access: agreement, authorized uses, retention limits, security controls, and responsible owner.
• Operations: delivery schedule or response target, outage behavior, retry rules, reconciliation, and staff fallback.
• Decision trace: the evidence and policy version used, any conflicting information, and the reason for the next action.

Maintain current information where permitted, but separate data maintenance from an eligibility determination. An incoming event should follow the applicable renewal or change-processing rules, notice requirements, and response protections. Continuous background collection is not authority for continuous adverse redetermination.

Pilot an integration change against representative cases and compare outcomes with a defined baseline. Examine false matches, missing records, manual work, source failures, and completed renewals. Faster retrieval is useful only when it improves the process without sacrificing accuracy or required protections.

Make the modernization backlog accountable

A procedural termination describes why a case closed; it does not by itself prove either ineligibility or an unlawful decision. Reviews need the underlying evidence, outreach history, required response period, and applicable rule. Likewise, a successful deployment or a closed corrective action is not a substitute for testing the next release.

CIOs and Medicaid leaders should maintain a joint backlog with a named owner, acceptance evidence, and an operational measure for each change. Keep distinct measures for ex parte completion, pending-case age, source failures, contact delivery, additional-information requests, staff processing time, and corrected decisions. Define denominators so that a falling backlog cannot hide cases that were closed incorrectly. Sample outcomes as well as counting transactions.

For the next policy change, use a readiness package that includes:

1. Versioned policy requirements: affected people, effective dates, required evidence, and decision rules approved by program owners.
2. Regression cases: individual-versus-household decisions, different eligibility categories, conflicting evidence, and incomplete submissions.
3. Operational proof: load and recovery results, staffing assumptions, source agreements, and escalation paths.
4. Resident-facing proof: understandable notices, working response channels, status visibility, and correction or appeal handoffs.
5. Post-launch review: a schedule for examining sampled decisions, exceptions, and unintended effects, with authority to correct defects.

These controls are useful across benefits programs, but each program needs its own legal and operational baseline. A reusable data service or queue does not make Medicaid, CHIP, and SNAP rules interchangeable. Build reusable technical capabilities while keeping policy decisions explicit and reviewable.

Sources and further reading

• CMS: May 2024 analysis of unwinding renewal completion timelines
• GAO: unwinding oversight findings and January 2026 recommendation update
• CMS: November 2024 ex parte renewal guidance
• CMS: January 2023 unwinding guidance, including contact-information conditions

Spartan X combines engineering, AI consulting, and program execution capabilities to help turn eligibility requirements into testable system behavior. That means connecting data and workflow decisions to acceptance evidence, practical operating plans, and a modernization backlog that program and technology leaders can manage together.