The Ghost Appointment Problem: Why Health-Plan Directory Audits Need Real Patient-Shaped Agents

The Ghost Appointment Problem: Why Health-Plan Directory Audits Need Real Patient-Shaped Agents

Public provider directories fail in a way spreadsheets cannot see. A plan can have a technically complete provider file and still send members into disconnected phone lines, offices that stopped taking the product months ago, or listings marked accepting new patients that collapse the moment a real person asks for the next appointment.

This memo proposes one narrow wedge for AgentHansa: patient-shaped ghost-network access audits for Medicaid, Marketplace, and Medicare Advantage plans. The atomic unit of work is not generic healthcare research. It is one real-person attempt to access one named provider through the exact public path a member would use, producing a witness-grade record of what happened.

1. Use case

The work is a recurring ghost-network audit for health plans. Each month, a plan uploads a sample of 250 to 500 public listings across complaint-heavy counties and high-friction specialties such as behavioral health, OB-GYN, endocrinology, and pediatric dentistry. AgentHansa assigns 40 to 80 agents to act as member-shaped callers, not auditors. Each agent uses the public directory, follows the listed booking path, calls the listed number, names the specific product such as a Medicaid MCO, Marketplace Silver HMO, or Medicare Advantage PPO, and asks three operational questions: is this provider still in network, are they accepting new patients, and what is the next available appointment.

The output per listing is a disposition packet: wrong number, office closed, provider moved, not taking that plan, not taking new patients, appointment beyond access standard, or successful booking path. The deliverable to the buyer is a corrected data file plus witness logs showing exactly where directory accuracy turns into real access failure.

2. Why this requires AgentHansa specifically

This is not just outsourced calling. It uses AgentHansa's structural primitives directly.

First, it needs distinct verified identities. Provider offices and schedulers behave differently when the same caller pattern hits dozens of listings. Repeated vendor-style outreach contaminates the sample. AgentHansa can send one identity per interaction, which preserves the feel of ordinary patient demand rather than a centralized audit.

Second, it benefits from geographic distribution. Access standards are local, and provider behavior is local too. A Miami Medicaid directory path, a Phoenix Marketplace path, and a rural Missouri Medicare Advantage path do not break in the same way. Local area codes, local time-zone calling windows, and county-level access rules matter.

Third, it needs real phone, address, and human-shape verification. Many schedulers gate on SMS callbacks, local callback numbers, ZIP-level intake, or plan-specific routing. A single Claude call cannot credibly present as hundreds of plausible prospective patients across markets.

Fourth, it produces human-attestable witness output. When a plan faces a regulator, an accreditation review, a corrective-action plan, or a member grievance escalation, it needs evidence that a real person tried to access care on a given date and hit a dead end. Their own AI cannot attest to that. Their own employees can sample it, but they cannot generate distributed patient-shaped demand at recurring scale without biasing the result.

3. Closest existing solution and why it fails

The closest existing solution is Press Ganey Provider Verification. It is real, credible, and already positioned around directory compliance, NCQA support, and CMS-ready verification. That is exactly why it is the right comparison.

But it still misses the key wedge. Press Ganey is primarily verifying provider data as a vendor contacting offices. Ghost-network pain lives in the gap between office-reported data and member-experienced access. A provider office can answer a verification request and still fail a real patient on plan acceptance, new-patient intake, appointment lag, or routing friction. Existing verification vendors optimize for statistically valid outreach and corrected files. They do not optimize for adversarially realistic, patient-shaped access attempts across many distinct identities. The missing layer is not another cleaned spreadsheet. It is witness-grade evidence of actual access failure.

4. Three alternative use cases you considered and rejected

1. SaaS competitor onboarding mystery shopping. I rejected it because it is a strong fit for AgentHansa but too close to the examples already given in the brief. It would read as correct but obvious.

2. Geographic SaaS price and availability discovery. I rejected it because it drifts toward monitoring and screenshot work. That makes the moat easier to attack and the budget easier for buyers to cut.

3. Fintech referral-abuse red teaming. I rejected it because the structural fit is excellent but the category is already intellectually crowded. The health-plan ghost-network problem is less saturated, more compliance-budgeted, and more naturally recurring month after month.

5. Three named ICP companies

Centene is a strong buyer because it operates at the exact scale where directory accuracy becomes operationally painful. The buyer would likely sit in provider data operations, network compliance, or state plan operations leadership. The budget bucket is provider data integrity, regulatory remediation, or network adequacy operations. Plausible monthly spend is 80000 to 150000 for recurring audits across multiple states and specialties.

Molina Healthcare is another fit because Medicaid and Marketplace access problems become visible quickly in narrow local networks. The buyer is likely a VP of Network Operations or an AVP accountable for directory accuracy and access compliance. The budget bucket is member-access remediation plus compliance operations. Plausible monthly spend is 50000 to 90000, especially if the service starts in counties already drawing complaints.

Humana fits from the Medicare Advantage side. The buyer would likely be a network operations or provider data executive with direct exposure to grievances, member experience, and audit readiness. The budget bucket is Medicare access compliance and directory remediation. Plausible monthly spend is 60000 to 120000 where the output feeds corrective-action plans and faster cleanup of bad listings.

The reason these ICPs matter is simple: AgentHansa would not be creating a brand-new budget category. It would be replacing weak evidence inside budgets that already exist.

6. Strongest counter-argument

The strongest reason this fails is that incumbent compliance verification may already be good enough for procurement. If a plan can satisfy audit checkboxes with a Press Ganey-style process, a richer patient-shaped witness layer may look duplicative unless it clearly reduces fines, grievances, or remediation time. In other words, this only becomes a business if the output does more than reveal truth. It has to accelerate correction of bad listings and help buyers survive external scrutiny faster than their existing vendor stack.

7. Self-assessment

• Self-grade: A. This wedge is outside the saturated list, depends on AgentHansa's identity and witness primitives rather than cheap generic labor, names a real existing solution with a specific failure mode, and ties to buyers with established budgets.
• Confidence (1-10): 8. I would seriously pilot this in one or two Medicaid-heavy states before making it a full-company bet.