The Sleeping Giant Won't Wake by Hiring Better Middlemen

Employer-sponsored insurance is the largest source of healthcare coverage in the United States, and most of it is self-funded, which means the employer (not an insurance company) is on the hook for the medical bills of its workers and their families. Self-funded employers are legally liable for every adjudicated claim, every appeal decision, and every dollar of out-of-network spread, while the information required to actually manage that liability sits inside the third-party administrator handling the plan. The employer carries the financial exposure, the entity it pays a per-member fee to handle the plumbing carries the information advantage, and provider organizations negotiate with the TPA on prices the employer never sees, in contracts the employer rarely reads, with carveouts and pass-throughs that the TPA may not be fully obligated to disclose. That is the structural reason employer healthcare costs have grown three times faster than wages for more than two decades, and it is the reason the strategies the industry has tried (narrow networks, higher cost-sharing, wellness programs, point solutions) have not bent the curve.

A recent NEJM Perspective by Zirui Song and Suhas Gondi correctly diagnosed this as the central reason self-insured employers, despite representing the largest purchasing block of healthcare in the country, have not moved the market. Where I would push the argument further is on the fix. The reforms the paper proposes (claims-data ownership, fee-schedule reweighting, contract terms against arbitrage) are all necessary, and they all have to be enforced by the same intermediary they are reforming. That is the central structural problem with the current path, and it shows up in three concrete ways.

Three places this shows up

Claims data. Federal law gives self-insured employers a right to their plan members' claims data, and federal price transparency rules support that right. In practice the data is TPA-formatted, TPA-keyed, sits in TPA systems, and arrives on TPA terms, which means quarterly file dumps with fields that matter (allowed amount, network discount math, contractual carveouts) populated inconsistently or not at all. Employers who hire analysts to make sense of these files end up reconstructing the adjudication trace from the bottom up, often without ever being able to validate the spreads being reported. The Affordable Care Act and the Consolidated Appropriations Act have both pushed the formal disclosure regime forward; the data infrastructure required to actually use those disclosures is something the TPAs would need to build for their employer clients, and most have not.

The fee schedule. Song and Gondi suggest that employers can use the price lever to pay more for mental health services or primary care and less for low-value services, encouraging clinicians into network. The mechanism for doing that runs through the TPA's adjudication engine, the TPA's network contracts, and the TPA's reporting. Most adjudication engines do not support surgical price adjustments. Most network contracts are written at the system level, where a single negotiated rate covers everything from primary care visits to spine surgery. Most reporting cannot tell an HR director, six months after a price change, whether utilization actually shifted or whether the additional dollars showed up as billing intensity in unrelated codes. The lever exists in policy, but the hardware to pull it sits behind a vendor whose business model is not aligned with employers pulling it.

Arbitrage protections. Recent Health Affairs reporting and several pending ERISA cases have documented TPAs burying administrative costs in claims, steering members toward their own subsidiaries (often at higher prices), and charging high "shared savings" fees on out-of-network claims that the TPA itself readjudicates at lower prices, sometimes leaving the worker subject to balance billing. Banning those practices contractually is necessary, but enforcing the ban requires claim-level transparency at the moment of adjudication, which is what the TPA controls. ERISA fiduciary suits work as a remedy years later, after the spread has been collected and the workers have been balance-billed, which is why the practice has continued even as litigation has accelerated.

What the post-TPA stack already looks like

Every meaningful reform of self-insured healthcare requires the employer to stop relying on the same intermediary for the data, the rules, and the audit. The infrastructure to do that is not theoretical. It is in operation today at a handful of large self-funded employers who built it because they had the scale to justify the engineering investment, and it is what the next decade of employer healthcare looks like at every size.

The first layer is direct provider contracting at the service-line level, owned by the employer or by coalitions of employers. Walmart's Centers of Excellence program, which contracts directly with a curated set of academic and integrated health systems for cardiac surgery, joint replacement, transplants, spine, and bariatric procedures, is the public version of this layer. Employees travel to the Centers, costs are bundled, outcomes are tracked end-to-end, and the TPA is not the entity that adjudicates those claims. The Purchaser Business Group on Health, a coalition of nearly forty self-funded employers covering roughly 21 million lives, is the broader version of the same idea: collective design at the technology and contracting layer, not just at the negotiating table. Both are real today, and both are constrained by the fact that the contracts they negotiate still get adjudicated, for the most part, by member TPAs whose payment engines were built for retrospective fee-for-service adjudication and not for outcome-based or service-line-weighted designs. The next generation of these coalitions has to bring its own adjudication.

The second layer is claims-data ownership at the source. Not a quarterly file from a TPA, but a real-time normalized feed of every claim, with the contracted rate, the allowed amount, the adjustment reasons, the spread between billed and paid, and a verifiable audit trail back to the provider's submission. This is software work, not negotiation work, and the vendors who build it are mostly new, the kind of companies that did not exist when the current TPA contracts were last renewed. The reason employers do not have this infrastructure yet is not that it is unbuildable, it is that the bandwidth required to procure, integrate, and operate it has historically not existed inside HR and benefits organizations structured around picking a carrier and signing a renewal.

The third layer is transparent adjudication. The software that decides whether a claim pays, and how, should be auditable in real time by the entity whose money is at stake. The current model, where the adjudication engine is owned by the firm earning fees on how it adjudicates, is a fiduciary inversion that the rest of the financial services industry would never tolerate; a bank's clearing system is not owned by the counterparty earning fees on the clearing. The fix is technically straightforward but institutionally hard, because the firms that benefit from the inversion are the same ones the policy debate keeps proposing to renegotiate with.

The federal answer is going to be one of two bad ones

The pressure on the current model is going to come from above before it comes from below, and there are two federal paths under active discussion that will accelerate the timeline. The first is the expansion of individual coverage health reimbursement arrangements (ICHRAs), which let employers convert their group health benefit into a defined contribution that workers use to buy coverage on the individual market. The mechanical effect is to shift the risk and the shopping back to the employee, into an individual market that is itself dominated by carriers running their own version of the same arbitrage. The second is taxing employer-sponsored coverage by capping the deduction or repealing the exclusion that currently makes employer contributions tax-deductible, a policy idea supported across the ideological spectrum at various points (the Affordable Care Act included the Cadillac tax, the Republican Study Committee has revived a version), which works by taking back the hundreds of billions of dollars annually in foregone federal revenue from a benefit that disproportionately accrues to higher-income workers.

Neither of those answers addresses the structural problem. ICHRAs offload it onto workers who have less information and less leverage than their employers do. Taxing employer coverage reduces the cost of the benefit to the federal government, not the cost of the underlying medical bills. Both paths leave the TPA stack intact and accelerate the timeline on which employers either move on their own or get moved on. The only path that actually addresses the cost curve is the one where employers take operational ownership of the data, the contracting, and the adjudication, and the window for doing that on their own terms, rather than under federal duress, is open right now and is closing fast.

What waking up actually looks like

A self-insured employer that acts on this in the next five years has an HR and benefits organization that looks materially different from today's. The team is smaller in headcount but deeper in expertise, including a claims data analyst who would have lived in an actuarial consultancy five years ago, an in-house clinical pharmacist who manages the pharmacy spend the PBM cannot or will not, and a benefits engineer (a role that does not yet have a standard title) who owns the integration between the claims feed, the provider directory, and the prior authorization rails. The TPA contract has been narrowed, not terminated. The TPA still handles run-of-the-mill claims adjudication on a transactional basis, but it does so on top of a data layer the employer owns, with contractual obligations to expose the adjudication trace in real time. Direct contracts with provider organizations handle a meaningful share of the high-volume, high-stakes procedures, and the local employer coalition is bidding the technology stack collectively. Per-employee spend growth has decoupled from premium growth in the rest of the market. Wage growth has resumed in the employee population. The benefit is more generous in the places that matter (mental health, primary care, complex care navigation) and tighter in the places that do not.

That picture is not a forecast or an aspiration. It is what the early movers already look like, and it is what the founders building the infrastructure layer for this market (myself included) are building toward. The reforms in the academic literature are useful but downstream of the actual thesis, which is that self-insured employers control the largest unmanaged purchasing block in American healthcare, and the next decade will be defined by whether they keep delegating that purchasing to entities whose business model depends on it growing. The employers who internalize that shift will spend the next decade on a different cost trajectory than the ones who do not, and the difference will show up first in worker wages and then in everything else.