Tucked inside every employer health plan's claims warehouse is a damning report nobody runs. Open a folder of last year's professional claims, lay it beside your TPA's current provider directory, and you'll find a gap most plan sponsors never see until an employee complains: the doctors your directory says are in-network, accepting patients, and ready to help, who haven't billed your plan a single dollar in twelve months. Some of them don't exist as accessible clinicians at all.
That gap is the ghost network, and it's far more than an inconvenience. For self-funded employers, it's a slow leak of trust, a compliance risk, and a hidden driver of downstream costs. The data to fix it is already sitting in your claims extract. The trick is knowing how to turn it into an audit of access rather than another cost-containment exercise.
Where Ghosts Come From and Why They Thrive
Provider directories are notoriously stale. The No Surprises Act requires plans to verify directory entries at least every 90 days, and CMS network adequacy rules impose their own accuracy standards, but enforcement is still catching up. In behavioral health, the trouble runs especially deep: a therapist who joined a group practice three years ago, moved on, but remains listed under the old NPI; a psychiatrist who stopped taking new patients during the pandemic and never updated their status; a group that dissolved but whose individual clinicians still float in the database. Employees search, call, hit voicemail, and eventually give up, or pay out of network. The plan sponsor rarely hears about it until renewal time, if at all.
From an ERISA standpoint, the threats are real. A plan fiduciary must act prudently and solely in the interest of participants. Knowingly presenting a directory full of false choices is the opposite of that. When those dead entries cluster in mental health, you're also flirting with a parity issue under MHPAEA: the plan may promise in-network behavioral coverage that's functionally nonexistent, while medical access holds up fine.
What Your Claims Data Knows That Your Directory Doesn't
Every time a member sees a professional, the claim (an 837 transaction, if you want to get technical) captures the rendering provider's NPI, taxonomy code, and place of service. By extracting a full year of professional claims where the provider was paid as in-network and deduplicating by unique member-provider encounters, you get an unvarnished "active contact" list. That's your ground truth. The directory file from your TPA is the promise. Align the two, and the fiction starts showing.
A sharp analytics team works through it in six steps:
- Build the ground-truth list. Export all professional in-network claims. Deduplicate to a set of NPI-member encounters. You now know exactly which providers rendered services.
- Get the directory in a usable format. Request the same file employees see through the portal. If your TPA sends a PDF or a clunky spreadsheet that won't join cleanly on NPI, that's your first warning signal.
- Match by NPI, never by name. Names are messy. NPI is the only reliable key. Pull both the billing and rendering NPI fields; a group practice often bills under its own NPI while the clinician who saw the patient carries a different one. Matching on one field alone flags real, busy providers as ghosts.
- Spot the cold spots. Pick a specialty like child psychiatry or endocrinology. Filter directory providers within a reasonable drive of your population centers. Cross-check with your claims list. Zero claims from any member over a year? That NPI is a ghost candidate.
- Measure access friction. Look at referral-to-visit lags for specialty care. When an initial primary care visit generates a referral, and the specialist claim shows out-of-network payment despite an in-network option appearing in the directory, the path was broken. The member likely hit a dead end.
- Overlay geography. Map employee ZIP codes to provider locations. Compare drive times to standard benchmarks: 30 minutes for primary care, 60 minutes for specialty. When the directory lists a dozen nearby behavioral health clinicians but only two ever bill your plan, employees are burning time and hope on the others.
From Insight to Contractual Teeth
Once you've identified clusters of ghost NPIs, the conversation with your TPA shifts from "We suspect there might be a problem" to "Here is the evidence, by specialty and by geography." That's the moment to insist on remediation and to write a quarterly audit into your administrative services agreement.
Something like this belongs in your ASA: "Administrator shall, on a quarterly basis, produce a cross-walk comparing the plan's provider directory to actual in-network claims utilization, disaggregated by specialty, and shall investigate and remediate any listing with zero paid claims and no evidence of active appointment availability within 30 business days." If that language makes your TPA blink, take note. They may know exactly how many ghosts are hiding.
This discipline also strengthens your stop-loss negotiation. A plan that can demonstrate it rigorously audits network access is a plan less likely to suffer the high-acuity claims that arise when care is delayed. For total rewards leaders, the story is even cleaner: when an employee can't find a therapist, they don't blame the directory; they blame the company. Audit the network, and you give people a reason to stay.
Enforcement and Lawsuits Are Already Underway
This isn't a problem regulators are waiting to study. In May 2023, Senate Finance Committee staff released a secret-shopper study of 120 mental health providers listed in 12 Medicare Advantage directories across six states; more than 80% turned out to be ghosts, and callers got an appointment only 18% of the time. States have moved from studies to penalties. In August 2025, New York's attorney general reached a $250,000 agreement with MVP Health Plan that requires the insurer to verify every listed mental health provider every 90 days, after a review found only 14% of 400 calls produced an in-network appointment. In February 2026, the same office secured a $2.5 million settlement with EmblemHealth, its largest for directory failures. A separate federal suit filed by the American Psychiatric Association and the New York State Psychiatric Association in December 2025 accuses EmblemHealth of padding its mental health directory with duplicates; one psychiatrist appeared 29 times.
Federal agencies are watching the same conduct. The 2025 MHPAEA report to Congress, issued in February 2026, shows the Department of Labor requested comparative analyses for 77 nonquantitative treatment limitations during the reporting period, with provider network admission standards the most common trigger. The agencies' enforcement guidance lists inaccurate provider directories as a leading parity barrier.
For a plan sponsor, this raises the cost of doing nothing. The audit in this post is now protection against the same conduct that has produced fines, restitution orders, and named-defendant status. Run the claims-to-directory cross-walk before a regulator or a plaintiff does, and you fix the directory on your own schedule instead of explaining it later.
Privacy Guardrails Are Non-Negotiable
None of this requires peeking into individual members' appointment attempts or call logs. That would cross into impermissible monitoring. The plan sponsor should receive only de-identified, aggregated data, provider-level rather than patient-level, as permitted under HIPAA. Work with legal counsel and an analytics partner who can deliver a population view without exposing protected health information. The goal is network integrity, not surveillance.
The Quiet Cost of Doing Nothing
Ghost networks don't just waste time. They breed distrust, force unnecessary out-of-network spending, and cause people to abandon care entirely when the search feels impossible. That's a cost that never appears on a stop-loss report but shows up in absenteeism, turnover, and a benefit that feels hollow. Your claims data already holds the most honest picture of the network your employees experience. You need to run the report.
One concrete step: Pull a twelve-month professional claims extract and your current directory file. Join them on NPI. Count the mental health providers with zero claims across your entire covered population. That number, all by itself, will tell you whether you're managing a network or an illusion.