Eligible but unreachable: Medicaid unwinding ran on contact data

The week

On April 1, 2023, states could begin removing people from Medicaid, the joint federal and state program that pays for health care for people with low incomes, for the first time since early 2020. For three years, once people were covered, they had stayed covered. During the COVID-19 public health emergency, states that accepted enhanced federal funding had to keep nearly all enrollees continuously covered. In normal times, states check eligibility regularly, usually once a year, and people whose income or circumstances have changed lose coverage. During the pause, those checks with consequences stopped.

The bargain dated from March 2020. The Families First Coronavirus Response Act raised the federal share of each state’s Medicaid spending by 6.2 percentage points, on condition that the state keep enrollees covered. States across the country took the money, and the condition stayed in force as long as the emergency did.

The pause ended by act of Congress. The Consolidated Appropriations Act, 2023, signed on December 29, 2022, ended that requirement on March 31, 2023, separately from the end of the emergency itself, which followed in May. It phased the extra funding down over the rest of the year: 5 percentage points from April to June, 2.5 from July to September and 1.5 from October to December.

The law attached strings to that tapering money. A state claiming it had to follow federal renewal rules and attempt to update each enrollee’s contact information before renewing them. It also had to try more than one way of reaching a person before ending coverage because a letter came back. States also had to report monthly data, including renewals, terminations, procedural terminations and call center wait times, through June 30, 2024. The Centers for Medicare & Medicaid Services (CMS), the federal agency that oversees Medicaid, would publish the figures and could demand corrective action plans from states that fell short.

The enforcement had teeth on paper. CMS guidance in January 2023 explained the sequence. A state that failed to submit or carry out a corrective action plan could be made to suspend procedural terminations, and could face civil money penalties of up to $100,000 a day. A state that drew the enhanced funding without meeting its conditions would have to give it back.

What followed was known as the unwinding. States had up to 12 months to start renewals for everyone enrolled and 14 months to finish them, and had to report outcome data to CMS. A renewal, or redetermination, is the process of confirming that someone still qualifies. States could start renewals as early as February 1, 2023, but no termination could take effect before April 1. CMS recommended that states start no more than one-ninth of their caseload in any month, to spread the work and set a pattern for future years.

States had prepared with federal help. By January 2023, CMS said, it had granted more than 40 states temporary waivers to use alternative renewal strategies. Examples included relying on SNAP income data and accepting address updates from health plans without first confirming them with the enrollee. SNAP is the Supplemental Nutrition Assistance Program.

A few states began terminating coverage in April; most started over the following months. Enrollment had grown substantially during the pause, and many enrollees had not been through a renewal in three years.

That last fact mattered more than any rule change. In three years, people move, change phone numbers, take new jobs and have children. Agencies were about to contact millions of people using records that, in many cases, no longer described their lives. Whether those people kept the coverage they were entitled to would depend on three things. A letter had to find them, or the state had to renew them without a letter at all, and someone had to answer the phone when they called.

The unwinding was the largest eligibility operation in Medicaid’s history. Its rules were mostly familiar: income limits, household composition, residency. What was unusual was the scale and the gap. Three years without renewals meant that addresses, phone numbers and circumstances had changed for many enrollees, and agency records had not kept pace.

Congress had also changed the incentives. By tying the last of the enhanced federal funding to renewal rules, contact updates and returned-mail handling, the Consolidated Appropriations Act turned operational process into a condition of payment. A state’s eligibility system settings and mailroom procedures were now, in effect, part of its federal financial compliance.

Outcomes depended on three distinct capabilities, which some states had and others lacked: renewing people automatically, reaching the people who had to respond, and handling the response once it came. They are matters of data, systems and operations more than policy, the kind that rarely reach a governor’s desk until something goes wrong. They apply to any organization that must re-verify customers at scale. The sections below explain the machinery and take each capability in turn. They then cover the measure that tied the three together, the costs that reached leadership, and how a well-run operation differed from a poor one.

What sits behind a Medicaid card

Behind every Medicaid card is an eligibility system: the software a state uses to record applications, check them against the rules and decide who qualifies. It holds each person’s details, such as name, address, household members and income, and it runs a rules engine, a component that applies the program’s eligibility rules to those details and produces a decision.

Modern eligibility systems also connect to other sources of information. They can query state wage records, other benefit programs and federal databases to confirm what an applicant has said, without asking them to send paperwork. It works the way a bank does when it confirms your salary directly with your employer rather than asking for three months of pay slips.

The quality of those connections, the accuracy of the underlying records and the way the rules are configured determine how much work falls on the agency and how much falls on the enrollee. When the system can confirm eligibility from data it already has, the enrollee does nothing. When it cannot, the enrollee gets paperwork, and every piece of paperwork is a chance for an eligible person to fall through the cracks.

Two kinds of enrollee, two kinds of check

Medicaid enrollees fall into two broad groups for eligibility purposes. Most children, parents, pregnant women and expansion adults are assessed on modified adjusted gross income, or MAGI, a measure based on federal tax rules. Their eligibility turns mainly on income and household size, which data sources can often confirm.

Older people and people with disabilities are often assessed under non-MAGI rules, which can include tests of assets such as bank balances. Asset checks are harder to automate. States use electronic asset verification systems that query financial institutions, but responses can be slow or incomplete. These enrollees, many of whom have stable circumstances and serious health needs, were often harder to renew without paperwork than younger adults with wage records.

Who runs the system matters

Eligibility systems vary widely in age and design. Some states run integrated systems that determine eligibility for Medicaid, SNAP and cash assistance together. Others run a Medicaid system that connects to the state’s health insurance marketplace. Many systems were built and are maintained by large contractors, so a configuration change can require a change order, testing and a release cycle measured in weeks or months.

Administration differs too. In some states, including California and New York, counties carry out much of the eligibility work under state rules. In others the state does it centrally. County administration can bring local knowledge and community ties, but it also means dozens of offices with different staffing, backlogs and practices, all feeding one set of statewide numbers.

Federal funding shapes what states can fix. CMS pays a 90 percent federal share for qualifying eligibility system development and 75 percent for ongoing operation, under approved plans. A state that found a defect in its renewal logic could usually get most of the fix paid for. The bottleneck was more often vendor capacity and testing time than money.

The unwinding was, at bottom, a test of how much each state could do on people’s behalf, and how well it could reach and serve them when it could not.

Renewing people automatically

Federal rules require states to try to renew eligibility from information they already have before asking an enrollee to fill out forms. Sources include state wage records, other benefit programs and the federal data services hub, a federal system that lets states check information such as income and citizenship against federal records. This is called ex parte renewal, a legal term meaning the decision is made without requiring the enrollee to take part. When it works, an eligible person stays covered without doing anything. When it fails, the person receives a renewal packet and must respond.

The sequence the rules require

The federal renewal regulation sets out a fixed sequence, and each step is a system function. The state must first attempt an ex parte renewal for every enrollee, MAGI and non-MAGI alike. If that fails, it must send a renewal form asking only for what it needs; for MAGI enrollees the form must be prefilled with what the state already knows. MAGI enrollees get at least 30 days to return it, and must be able to respond online, by phone, by mail or in person.

Before ending coverage, the state must consider every basis on which the person might qualify, including bases other than the one under which they were enrolled. It must give at least 10 days’ notice and fair hearing rights. If a MAGI enrollee loses coverage for failing to return the form and then sends it within 90 days, the state must reconsider without requiring a new application. Each of those steps depends on correct configuration, and a gap in any one produces avoidable losses.

Why ex parte rates differed

Ex parte rates varied widely across states. They depended on whether the eligibility system could match records reliably across programs. They also depended on how many data sources it could query in a batch, meaning in one large automated run rather than case by case. They also depended on how the rules engine handled edge cases such as mixed households, irregular income and people with no recent wage records.

Small rules decisions moved large numbers. States set a reasonable compatibility threshold: if the income in the data differs from the income on file by less than a set margin, the state can treat them as consistent. A tight threshold sends more people to paperwork. People with no income in any data source posed a similar choice. A state could renew them on the basis that no wages were found, or demand proof of zero income, which few people can produce. CMS waivers allowed several states to take the first path.

States with integrated eligibility systems for Medicaid, SNAP and other programs had an advantage, because income verified for one program could support another. SNAP is the Supplemental Nutrition Assistance Program, often still called food stamps. If a family’s income was confirmed for food assistance last month, an integrated system can use that to renew their health coverage too. States whose Medicaid systems had few data connections mailed far more packets.

The hidden switch: individuals versus households

The configuration detail that mattered most turned out to be the unit of renewal. Eligibility is determined person by person, but many systems processed renewals by household. If one adult’s income could not be verified, the whole household went to a packet, including children who qualified on data the state already held.

Take a family of two parents and three children. The children qualify easily; the state’s records confirm it. One parent has irregular self-employment income that the data sources cannot verify. In a system that renews by household, the whole family receives a packet. If the packet goes to an old address, all five lose coverage, including three children the state already knew were eligible. In a system that renews by person, the three children are renewed automatically and only the parent receives a request for information.

The same flaw appeared in a second pattern. In a mixed household, one member might need to supply a document the others did not, such as proof of immigration status. A household-level system held everyone’s renewal until that one document arrived. People whose own eligibility was clear lost coverage over a relative’s paperwork.

That design choice, invisible to leadership, drove avoidable coverage losses. Nobody debated it in public: it was a system setting, chosen years earlier, perhaps for reasons of convenience.

Testing the configuration before the first batch

A director could have found the switch with a test that takes days, not months. The team builds a set of synthetic cases covering the hard patterns. They include mixed households, a child with an adult who cannot be verified, zero-income adults, non-MAGI enrollees with assets and people with income near the threshold. It runs them through the renewal process in a test environment and checks each outcome against what the federal rules require. Every case that lands in a packet when it should have renewed points to a rule to fix.

The same test set should run again after every system release. Fixes made for one problem can reopen another, and the release notes from a vendor rarely say so.

A weekly count by county

An agency director needed operational answers. What share of renewals completes ex parte, by eligibility group and by county? Which data sources are queried, and which fail most often? Are rules applied to individuals or to households? A weekly dashboard of those numbers would have shown the problem early. A dashboard is simply a regularly updated summary of key figures, ideally on a single page. Broken down by group and county, it shows where the program is failing and for whom, which an overall figure hides.

The data source failure rates deserve their own line. If the wage database returns no match for a large share of queries, the problem may be identity matching, such as name formats or missing Social Security numbers, rather than a lack of income. Fixing the match logic can move thousands of people from packets to automatic renewal.

Reaching the people who had to respond

For enrollees not renewed automatically, the process depended on a letter reaching them and a reply arriving by a deadline. After three years, many mailing addresses were out of date. People who never received a packet could lose coverage for procedural reasons. That meant losing coverage not because the state found them ineligible, but because it could not confirm eligibility.

This distinction is the heart of the matter. Someone cut off for procedural reasons may still be exactly who the program is for. They simply did not complete a step, often because they never knew it was required.

What the law required before a termination

The appropriations law set two contact conditions for states claiming the tapering funds. Before renewing someone, a state had to attempt to ensure it had current contact information. The sources named were the Postal Service’s National Change of Address database, records held by state health and human services agencies, and other reliable sources. CMS read that as covering mailing address, phone number and email, and said a quarterly match against the postal database would count as a recent attempt.

The second condition applied when a renewal letter came back undelivered. The state had to make a good-faith effort to reach the person in more than one way, such as phone, email, text or an online portal, before ending coverage on that basis. CMS guidance distinguished three kinds of returned mail: with an in-state forwarding address, with an out-of-state forwarding address, and with no forwarding address. Each needed its own workflow. An out-of-state forwarding address, for example, raises a residency question, but it is evidence to check rather than a verdict.

The tools for finding people

States had several tools to improve contact data. Federal guidance allowed them to update addresses using Postal Service change-of-address data, managed care plan records and other programs’ files. Temporary waivers let many states accept those updates without first confirming them with the enrollee, which removed a step that would itself have required a letter.

Many states partnered with managed care organizations, which often held more current phone numbers and emails, to remind members. Managed care organizations are the private health plans that many states pay to deliver Medicaid benefits. Because they process claims and contact members regularly, they often know where people are better than the state does. Text messaging reached people who had moved but kept the same phone.

Some populations needed different routes. CMS urged states to work with the Indian Health Service, tribes and tribal organizations to reach tribal members, given limited postal delivery and broadband in some tribal communities. People without stable housing, rural residents and people with limited English needed channels such as clinics, food banks and community health workers, and notices in their own language.

The groundwork behind the tools

Each tool depended on groundwork. Agencies needed consent to text, data-sharing agreements with plans, and the technical ability to load updated contacts into the eligibility system. A data-sharing agreement is the legal arrangement that lets two organizations exchange personal information for a defined purpose. Without that groundwork, updated information sat in a spreadsheet. A health plan might know a member’s new address, and the state might even receive it. If nobody could load it into the system that generated the renewal letter, the letter still went to the old address.

Calls and texts carried their own legal checks. CMS reminded states that outreach by phone and text, whether by the agency or a contractor, had to comply with federal communications law, including the Telephone Consumer Protection Act. The Federal Communications Commission issued guidance to make such calls easier. Agencies still had to confirm consent and opt-out handling before launching a texting campaign at scale.

Data sharing also had limits. Medicaid confidentiality rules restrict what a state can share and with whom. A sound agreement with a plan or community partner names the purpose, the data fields, the retention period and the security controls. Drafting it takes legal time, which is why the states that had agreements in place before 2023 moved fastest.

One message, many messengers

Message consistency mattered too. Enrollees who heard different instructions from the state, their health plan and community groups were more likely to wait or give up. If the state says “watch your mail,” the health plan says “call us,” and a community group says “go online,” a busy parent may reasonably decide to wait until it becomes clear.

The agencies that did best agreed on one set of messages with their partners and timed outreach to the renewal calendar. They also counted returned mail and acted on it. Every envelope that came back undelivered was a signal that a person was about to lose coverage for procedural reasons, and a prompt to try another channel.

Timing mattered as much as wording. An outreach text sent two weeks before the renewal packet primes people to look for it. The same text sent after the deadline only generates calls about coverage already lost. Building outreach into the renewal schedule, month by month and cohort by cohort, turns partners’ goodwill into results.

Handling the response

The third capability was processing volume. Renewal packets went out in monthly batches that were large compared with normal operations. Enrollees who responded did so through online portals, call centers, mail and in-person offices. Each channel had capacity limits.

Call centers were the most visible pressure point. Long hold times and dropped calls meant people trying to comply could not complete renewal before the deadline. Portals that were hard to use on a phone pushed more people toward the phones. A portal is the website where people can log in and manage their case. Many low-income households rely on a phone rather than a computer for internet access, so a portal that works poorly on a phone works poorly for them.

Congress made those pressure points public. The monthly reports states had to file included total call volume, average wait time and average abandonment rate for each state call center. Abandonment is the share of callers who hang up before reaching anyone. Publishing those numbers state by state meant a long queue was no longer an internal problem.

Paper documents mailed in had to be scanned, indexed and matched to a case by staff who were also handling new applications. Indexing means labeling each document so it can be attached to the right person’s file. Backlogs in any of these steps produced procedural terminations of people who had in fact responded. Someone could do everything right, mail their pay stubs on time, and still lose coverage because the envelope sat unopened in a mailroom when the deadline passed.

The 90-day window as a safety net

The 90-day reconsideration period gave agencies a second chance, if they used it. A MAGI enrollee who lost coverage procedurally and then returned the form within 90 days had to be reconsidered without a new application. That requires a system that can reopen a closed case, staff who know the rule, and notices that tell people the window exists.

A state that handled reconsiderations as new applications imposed extra work on itself and longer gaps on enrollees. A state that tracked reconsiderations as a separate queue could see how many procedural closures were reversed, and how quickly. That number is a direct measure of how many eligible people the process had wrongly cut off.

Planning for a surge

Capacity planning for this kind of surge is well understood in other sectors. Retailers plan for the holidays; tax agencies plan for filing season. The steps are familiar:

• Forecast volume from the renewal calendar. The agency knows how many packets go out each month, so it can estimate calls, uploads and mail.
• Staff and contract for peaks. Bring in temporary or contracted capacity before the wave arrives, not after.
• Offer callbacks instead of holds. Let callers leave a number and be called back, rather than wait or give up.
• Make the portal work on mobile devices, with document upload by phone camera, so people can photograph a pay stub rather than mail it.
• Measure time from receipt to processing, as well as call answer times. A call answered quickly does not help if the document it concerns sits untouched for weeks.

The one-ninth guideline helped here. Spreading renewals evenly across the unwinding produced a predictable monthly load that staffing plans could match. States that front-loaded renewals to clear the caseload quickly took on bigger peaks, with more risk of backlog-driven terminations.

States also used federal flexibilities to extend deadlines or pause terminations where backlogs built up. That was a reasonable choice, but it stretched the timeline and the workload. Every month of delay meant another month of overlapping cohorts, with new packets arriving while old ones were still being processed.

Where people go when they no longer qualify

Handling the response also meant handling the people who truly no longer qualified. Federal rules require a state that finds someone ineligible to assess them for other affordable coverage, such as CHIP or subsidized marketplace plans, and transfer their account. CMS told states using the federal marketplace platform not to transfer accounts of people terminated for procedural reasons, because the state had not actually determined them ineligible.

That rule made the quality of the determination matter twice. A person correctly found over the income limit could be handed to the marketplace with their information. A person cut off for a missing form left with nothing transferred, and had to start again on their own.

The number that tells the truth

Across all three capabilities, the measure that mattered was the share of people losing coverage for procedural reasons. A high rate means the agency could not complete the process. It does not mean those people were ineligible.

Many returned to coverage within months, after a gap that cost them care and cost the agency a new application. This churning, as it is called, is expensive for everyone. The enrollee may skip a prescription or delay a doctor’s visit during the gap. The health system may end up treating them later, sicker, in an emergency room. The agency processes the same person twice. And public trust in the program erodes each time someone is cut off for paperwork rather than eligibility.

Procedural terminations are the clearest signal of whether systems and contact data are working. A rising rate in a particular county, or among children, or among people renewed by mail, points directly to where the operation is breaking down.

Leading and lagging indicators

The procedural rate is a lagging indicator: by the time it rises, people have already lost coverage. A director needs leading indicators that move first. The ex parte rate shows how much work the system is doing on people’s behalf. The returned-mail rate shows how much of the contact data is stale. The response rate on renewal packets shows whether outreach is landing. Time from receipt to processing shows whether the back office is keeping up.

After the fact, two more numbers complete the picture. The reconsideration rate shows how many procedural closures were reversed within 90 days. The reapplication rate, people who return within a few months as new applicants, measures churn directly. Together they separate people who truly left the program from people the process pushed out.

Reading the number honestly

The procedural rate needs care. Some people terminated for procedural reasons were in fact no longer eligible: they had taken jobs with employer coverage or moved out of state, and saw no reason to reply. That is the strongest objection to treating the procedural rate as a measure of failure.

It does not overturn the argument. A procedural termination means the state does not know whether the person was eligible. The reconsideration and reapplication rates are how a state finds out. Where many procedurally terminated people come back within months, the process failed eligible people. Where few return, the procedural rate may overstate the problem. A state that tracks both can tell the difference, and a state that tracks neither is guessing.

The costs that reach the director

Each capability connects directly to things leaders are held accountable for.

Money. Every eligible person who loses coverage for procedural reasons and later reapplies generates a second round of work: a new application, new verification, new letters. Paying for that work twice is a cost the agency chose, usually without realizing it, when it decided not to invest in better data connections or more processing capacity.

The appropriations law added a direct financial exposure. A state drawing the enhanced funding was, by doing so, attesting that it met the conditions, subject to later review. If CMS later found it had not, the state would have to return the extra federal money for each quarter it failed to qualify. A configuration flaw could therefore become a budget item.

Risk. Configuration choices that conflict with federal requirements can become compliance problems. A system setting that seemed like a technical convenience can turn into a finding that requires corrective action, paused terminations and reinstatements, all under public scrutiny.

The corrective action timetable was tight. Under the law, a state notified of noncompliance had 14 days to submit a plan. CMS had to approve it within 21 days of submission, and the state had to begin carrying it out within 14 days of approval. A team that could not quickly diagnose its own system would struggle to meet those deadlines.

People. Caseworkers and call center staff bear the strain of a poorly prepared surge. Long queues mean angry callers, overtime and burnout, at exactly the moment the agency most needs experienced staff to stay.

Partners. Hospitals, clinics and health plans feel the effects when patients arrive without active coverage. They become, in effect, unpaid outreach workers, and they will judge the agency by how well it coordinated with them.

Each partner type feels it differently. A health plan loses its monthly payment for every member who drops off, so its incentive to help members renew is strong and well aligned with the state’s. A hospital sees the gap when an uninsured patient arrives for care and may screen them for presumptive eligibility, a temporary coverage route many hospitals can use. A community health center may spend staff time helping patients with forms that the state could have avoided sending.

Trust and reputation. Stories of children losing coverage despite the state holding proof of their eligibility are the kind that reach newspapers and legislative hearings. An agency that can show its ex parte rate, its procedural termination rate and what it is doing about both is in a far stronger position than one that cannot.

The same logic applies well beyond Medicaid. Any organization that must periodically re-verify large numbers of customers, from insurers to utilities to universities, faces the same three tests. It must confirm people without asking them, reach those it must ask, and handle their answers on time.

Banks run a version of this in periodic customer due diligence refreshes, and insurers in policy renewals. In each case the cost of an avoidable lapse falls partly on the institution, as rework and lost relationships, and partly on the customer. The design questions are the same: how much can be confirmed from data already held, how current the contact records are, and whether the back office can absorb a peak.

What a well-run unwinding looked like

Suppose a state agency set out to do it well. Before the first packet goes out, it checks that its rules engine renews each person individually and adds data sources where match rates are weak. It signs data-sharing agreements with its managed care plans and builds the pipeline to load their contact updates directly into the eligibility system. It agrees one set of messages with plans and community partners. It forecasts monthly volume, contracts extra call center staff for peak months, offers callbacks and lets people upload documents by phone. Every week, its director sees a one-page dashboard: ex parte rate, procedural termination rate and processing times, broken down by group and county.

A sequence, not a checklist

The order of those steps matters. Configuration testing and data agreements take the longest, so they come first, months before renewals begin. Contact updates should run shortly before each cohort’s renewal, so the address is fresh when the packet is printed. Outreach follows the packet calendar. Staffing peaks should match the response peaks, which arrive a few weeks after each mailing.

Governance holds it together. A small unwinding command group, with the Medicaid director, the eligibility system owner, the operations lead, a data analyst and a partner liaison, meets weekly on the dashboard. It has authority to pause a cohort, extend a deadline or push a system fix ahead of other work. Without that authority, problems found on the dashboard wait in queues.

The trade-offs a director faced

Every choice had a cost. Moving fast reduced the time the state paid for people who no longer qualified, which mattered to state budgets as the enhanced funding fell. It also raised the risk of backlog-driven terminations. Moving slowly protected eligible people but prolonged overlapping workloads and staff strain.

Program integrity and coverage retention pulled in different directions too. Legislators and auditors expected ineligible people to be removed. Advocates and providers expected eligible people to stay. Both expectations were legitimate, and the measures above are what let a director show that the state was meeting both.

The ways it went wrong were just as recognizable:

• Not checking the configuration. Assuming the system renews individuals when it actually renews households.
• Collecting contact data you cannot use. Receiving updated addresses from plans but having no way to load them.
• Mixed messages. Letting the state, plans and partners each give different instructions.
• Measuring the wrong thing. Watching call answer times but not how long documents wait to be processed.
• Ignoring returned mail. Treating undelivered packets as a postal issue rather than an early warning.
• Reading terminations as ineligibility. Assuming that people who lost coverage did not qualify, when many simply never completed a step.

Where procedural losses begin

Any agency facing a renewal surge can test its readiness with five questions, each tied to one of the capabilities above:

1. What share of our renewals complete automatically, and how does that vary by eligibility group and county?
2. Does our system renew each person individually, or can one person’s missing information send a whole household to paperwork?
3. How current are our addresses and phone numbers, and how many renewal letters are coming back undelivered?
4. How long does a document take from arriving in our mailroom to being processed?
5. What share of people losing coverage are losing it for procedural reasons, and is that number rising anywhere?

With hindsight

The household-level setting turned out to be the unwinding’s defining technical failure. In late August 2023 CMS warned states that running automated renewals at the household level, rather than for each person, violated federal requirements. Dozens of states paused affected procedural terminations, and reports put the number reinstated at close to half a million people. A configuration choice that few leaders had known about had become a matter of federal enforcement.

The scale of the problem was visible in CMS’s own summary. As of September 21, 2023, according to CMS, 23 states and territories attested that they were renewing at the individual level. Another 30, including the District of Columbia, said they were not or were still reinstating people. The affected groups were the two patterns described above: children in households with an enrolled adult, and household members held up by another member’s missing documents.

The overall numbers confirmed how much the operation turned on process rather than eligibility. By the time KFF, a health policy research organization, stopped updating its tracker in September 2024, about 25 million people had been disenrolled, roughly 69 percent of them for procedural reasons. That means most people who lost coverage did so because the process was not completed, not because the state had determined they no longer qualified.

The same tracker showed the other side. About 56 million people had been renewed, and about 61 percent of those renewals were completed ex parte, without the enrollee having to return a form. Automatic renewal carried most of the load, which is why a flaw in how it was configured mattered so much.

The unwinding was largely complete by mid-2024, and CMS rules finalized that year streamlined enrollment and renewals.

Some of the unwinding’s structure did not last. The monthly public reporting Congress required ran through June 30, 2024, and the special enforcement tools were tied to the same window. When the surge ended, so did much of the outside visibility into how individual states were performing.

The 2025 budget reconciliation law then required six-month redeterminations for many expansion adults and work requirements that KFF described as starting in January 2027. Expansion adults are the working-age adults who became eligible when states expanded Medicaid under the Affordable Care Act. Checking eligibility twice a year instead of once, and adding a new requirement to verify, means more renewals, more contacts and more documents.

Enrollment had already been falling before those rules took effect. KFF’s tracker, using preliminary data, put Medicaid enrollment at about 73 million in June 2026, roughly 2 percent above its February 2020 level and down about 5 million from a year earlier. The pandemic-era growth had largely reversed.

That turns the same three capabilities from a one-time surge into a recurring operation. Renewing people automatically, keeping contact data current and processing responses quickly are no longer emergency projects; they are the permanent machinery of the program. The investments made, or not made, during the unwinding will now be tested again, on a regular schedule, and the procedural termination rate will keep telling them how well that machinery works.

Work requirements add a fourth kind of check. Verifying hours worked, school enrollment or an exemption draws on data sources that Medicaid systems were not built around, and exemptions for groups such as caregivers or people with disabilities need their own evidence. States that learned in 2023 to test configurations before the first batch, to load partner contact data and to watch procedural rates by group start with an advantage.

What held up was the argument that data, not rules, decided outcomes. What did not was the hope that the unwinding was a single event to get through. By 2026 it looked more like a rehearsal.