Fragmented Employment and the ESI Assumption: Why the Coverage System Breaks When the Employment Unit Shrinks

LFP-12.04 | Sharp Analysis | Series 12: The AI Disruption

The employer-sponsored insurance system was built on assumptions about employment that were empirically reasonable in the postwar decades when group health coverage became the dominant form of private insurance in the United States. Health insurance became attached to employment primarily because of wage controls during World War II, tax treatment of employer contributions, and the administrative logic of pooling risk across groups of workers. The system never required a policy decision that employment was the right vehicle for health coverage. It required only that most workers have stable, full-time employment relationships with a single employer who had enough employees to form a viable risk pool.

Those conditions are deteriorating. They were deteriorating before AI arrived. AI is accelerating the deterioration by dissolving the employment units the system requires at a rate that exceeds prior structural trends.

The Three Embedded Assumptions

#

Employer-sponsored insurance embeds three structural assumptions about the employment relationship. Identifying them precisely matters because each is being undermined by a different mechanism, and the policy and product responses differ by mechanism.

The first assumption is that each worker has a single primary employer. The system assigns coverage to the employer-employee dyad. An employee is covered by their employer. A worker serving three clients as a fractional contractor has no primary employer and falls outside the system. The ERISA framework that governs employer plans does not contemplate multi-employer contribution arrangements for single workers. The worker who generates income across multiple client relationships must obtain coverage from a source entirely separate from any of those clients.

The second assumption is that the employer has enough employees to form a viable risk pool. Self-funded plans require actuarial stability that cannot be achieved below a minimum group size. Stop loss underwriting below 10 lives approaches individual insurance pricing, as analyzed in LFP-02.08. The employer with three to five employees cannot maintain a level funded or self-funded plan that is economically superior to fully insured coverage or individual insurance, because the variance in expected claims is too high relative to the expected claims level for the pooling mechanism to create value. The ESI model assumes employers are large enough to benefit from pooling. For employers below roughly 5 or 6 employees, the assumption does not hold.

The third assumption is that the employment relationship is stable enough to support an annual plan year. Plan documents, benefit elections, stop loss policies, and administrative processes are structured around the plan year. A worker whose client engagements last three to six months cannot sustain this structure through any single employer relationship. The fractional worker who cycles through three client relationships in a year does not have an employment relationship stable enough to anchor an annual plan.

The Pre-AI Erosion Baseline

#

The erosion of these assumptions predates AI. ESI coverage as a share of the nonelderly population declined from 67% in 1999 to 56% in 2014, according to KFF analysis of National Health Interview Survey data. The Peterson-KFF Health System Tracker documented a nine percentage-point decline in the ESI share of the nonelderly population between 1998 and 2018. ESI coverage as measured by the Current Population Survey Annual Social and Economic Supplement has held roughly steady in the 53% to 55% range since the ACA marketplaces opened in 2014, stabilized partly by the individual mandate and partly by expanded Medicaid (Peterson-KFF Health System Tracker; KFF, “Trends in Employer-Sponsored Insurance Offer and Coverage Rates, 1999-2014”).

The pre-AI erosion was driven primarily by cost. Employer offer rates declined as premiums rose faster than wages; employee take-up rates declined as employee premium contribution requirements increased. The structural driver was affordability within existing employment relationships, not the dissolution of those relationships. This is the diagnostic distinction that matters. Cost-driven erosion is addressable through subsidies, mandates, or cost reduction. It does not require new product categories or regulatory frameworks. The employment relationship still exists; the question is whether the parties can afford to fund it.

AI-driven erosion is categorically different. The employment relationship itself stops existing. The employer does not stop offering coverage. The employer’s headcount falls below the viable threshold, or the employer converts workers to independent contractor arrangements, or the employer restructures from a 20-person firm to a 10-person firm with the remaining 10 workers reorganized as fractional operators serving multiple clients. In each case, the coverage gap is structural: there is no employment relationship from which coverage could be offered, regardless of subsidy or mandate.

Worker Classification and Coverage Eligibility

#

Worker classification law is the regulatory mechanism that determines which workers fall inside and outside the ESI system. An employee is eligible for employer group coverage. An independent contractor is not eligible for employer coverage from the client, regardless of how substantial or ongoing the relationship is.

The DOL’s independent contractor rule has cycled through three iterations in five years. The Biden administration issued a final rule effective March 11, 2024, adopting a six-factor totality-of-circumstances test that makes it harder for businesses to classify workers as independent contractors under the Fair Labor Standards Act. The six factors examined are: opportunity for profit or loss based on managerial skill; investments by the worker and employer; permanence of the working relationship; nature and degree of control over the work; whether the work is integral to the employer’s business; and the worker’s skill and initiative. The rule treats economic dependence on the employer as the ultimate inquiry (U.S. Department of Labor, “Employee or Independent Contractor Classification Under the Fair Labor Standards Act,” 89 Fed. Reg. 1638, Jan. 10, 2024). By February 2026, the Trump administration had proposed rescinding the 2024 rule and returning toward the more employer-friendly 2021 framework, signaling continued regulatory volatility in this area.

California’s AB5, codifying the ABC test for worker classification, went in the opposite direction: it creates a presumption of employee status that must be overcome by demonstrating that the worker performs work outside the usual course of the hiring entity’s business, operates an independently established trade or occupation, and is free from control in the performance of the work. The Supreme Court of California upheld Proposition 22 in July 2024, which carves out app-based transportation workers from AB5’s strictest requirements, creating further classification complexity within the state.

The practical consequence is a patchwork: classification standards vary by state, federal law, purpose (FLSA versus IRS versus ERISA), and administration. The fractional professional serving multiple clients may be an independent contractor under federal FLSA standards, an employee under California’s ABC test, and something indeterminate under the IRS tests used for tax reporting purposes. This classification ambiguity does not create coverage eligibility. It creates compliance risk for the clients and structural coverage uncertainty for the worker.

The Pace Mismatch

#

The employment structure is changing at a pace the coverage infrastructure cannot match. Regulatory frameworks respond to workforce patterns that are already established and politically organized. Portable benefits legislation that would allow employers to contribute to worker benefits accounts regardless of employment classification, sponsored by multiple federal legislators including Senator Mark Warner, has been in proposal or pilot stages for years without federal enactment. The Aspen Institute’s Financial Security Program has documented the legislative proposals and the gaps they are designed to address. States including Washington and New Jersey have enacted limited portable benefits pilot programs, but none has produced a scalable model that addresses the full range of benefits that ESI provides.

Multi-employer contribution mechanisms that would allow multiple clients of a fractional worker to collectively fund benefits eligibility do not exist at scale under ERISA. ERISA’s framework for multi-employer plans was designed for collectively bargained arrangements in specific industries, not for the fluid, variable multi-client relationships that the AI-augmented fractional workforce generates.

The regulatory frameworks that would address the structural coverage gap for fragmented workers are in early stages of proposal or debate. The workforce that needs them is growing now. High-propensity business applications, the Census Bureau’s measure of EIN filings most likely to produce employer firms, have run above pre-pandemic baselines through 2024 and 2025, concentrated in professional services and information industries: the categories most exposed to AI-driven employment restructuring (U.S. Census Bureau, Business Formation Statistics). The micro-employer and fractional professional populations are expanding at a pace that the regulatory and product response has not matched.

The Level Funded Position

#

Level funded is positioned at a specific intersection in this structural shift. It serves the 1-to-50 employer market, where the fragmentation dynamic arrives first and most acutely. It operates through ERISA, which provides the federal preemption shield that makes self-funded plans viable regardless of state-level regulations. It has the product flexibility to adapt, at least architecturally, in ways that other coverage vehicles do not.

The limitation is not architectural. It is actuarial and administrative. The stop loss underwriting barrier below 10 lives is a constraint imposed by the math of variance, not by product design decisions that can be changed. The administrative cost structure of plan onboarding and compliance does not scale efficiently to very small groups. These constraints are real and are not eliminated by acknowledging the coverage gap.

What level funded’s position in this market does provide is proximity to the problem. The TPA serving 30-person construction companies, 15-person professional services firms, and 20-person regional distributors is watching those groups shrink toward viability margins in real time. The cost-driven erosion that preceded AI required policy responses: subsidies, mandates, and premium tax credits. The structural erosion that AI is producing requires product innovation: new pooling mechanisms, simplified underwriting models, and administrative structures that change the fixed-cost architecture of small group plan administration.

Whether those product innovations arrive before the structural erosion makes the problem insurmountable for level funded’s addressable market is the question this series cannot answer with certainty. The trajectory of the workforce change is clear. The pace of the product response is not.