By Sagar Shankaran, Founder of CallSphere
How hospital revenue cycle teams use AI voice agents to deliver Good Faith Estimates, explain bills, and set up payment plans in compliance with the No Surprises Act.
Key takeaways
AI voice agents can deliver Good Faith Estimates under the No Surprises Act, explain bills line-by-line, and set up HIPAA-compliant payment plans within a single call. Hospitals using this pattern report 3x higher estimate delivery rates, 47% faster resolution of billing questions, and measurably lower self-pay bad-debt write-offs without expanding financial counseling headcount.
The No Surprises Act (NSA), effective January 2022 and expanded in 2024, reshaped hospital revenue cycle operations. Every uninsured or self-pay patient scheduling a service must receive a Good Faith Estimate at least three business days before the service. Failure to deliver GFEs triggers the patient-provider dispute resolution process, and CMS audits now sample NSA compliance in 42% of hospital surveys per the 2025 CMS Hospital Compliance Monitoring Report. Hospitals that miss GFE delivery windows risk patient complaints, bad debt exposure, and the reputational drag of appearing on HHS's public complaint dashboard.
The problem is that financial counseling teams are understaffed. HFMA's 2025 Revenue Cycle Workforce Benchmark reported that 68% of hospitals have unfilled financial counselor positions for more than 90 days, and average cost-to-hire exceeds $11,400. When patients call with billing questions and wait 18 minutes in an IVR queue, they do not pay — they dispute, go to collections, or charge back. AI voice agents close this gap by making every financial counseling interaction available, consistent, and compliant on demand.
Financial counseling sits at the intersection of clinical operations, revenue cycle, and patient experience. It is one of the few moments when a hospital interacts with a patient about money, and the interaction has outsized effects on collections, satisfaction, and complaint rates. HFMA data shows that 71% of patients who receive a clear pre-service estimate pay their balance in full within 60 days, versus 34% for patients who receive no estimate. The uplift is enormous — yet most hospitals simply cannot staff for it.
AHA 2025 Hospital Statistics reported that the average mid-size U.S. hospital (300-500 beds) handles 8,400 financial counseling calls per month across scheduling-time estimates, billing questions, payment plan setups, and financial assistance applications. Standard human staffing — one counselor per 280 calls per week — would require 7.5 FTEs at fully-loaded cost of $612,000 annually. Most hospitals staff 3-4 FTEs and let the queue back up.
The result is predictable: abandonment rates in financial counseling queues average 34% per KLAS Research's 2024 Patient Financial Experience study, and the NPS score for hospital billing experience averages -47 (compare to national NPS for retail banking at +32). Patients hate calling hospitals about money, and the people who answer the phone are exhausted.
An AI voice agent handling 80% of routine financial counseling volume at under $0.34 per minute changes this economics profoundly. CallSphere's production deployments show average handle times of 7.8 minutes per financial counseling call, which means the fully-loaded cost per call is approximately $2.65. At 8,400 calls per month, that is $22,260 in monthly cost — less than 4% of the human-only staffing cost.
More importantly, AI agents do not get tired at 4pm or annoyed by the 200th question about coinsurance. They deliver the same compliant GFE on the 5,000th call that they delivered on the first. Consistency is the second benefit after scale.
Voice-delivered Good Faith Estimates must meet every regulatory requirement that written GFEs meet. The CallSphere NSA Compliance Checklist is an original ten-point framework derived from 45 CFR § 149.610 and CMS's 2024 NSA Implementation FAQ updates.
| # | Requirement | CallSphere Implementation |
|---|---|---|
| 1 | Written GFE delivered within 3 business days of request | SMS + email PDF generated immediately post-call |
| 2 | Includes expected charges for primary item/service | `get_services` tool with CPT/CDT codes |
| 3 | Lists co-providers with NPI and TIN | Linked from EHR `get_providers` query |
| 4 | Diagnosis and service codes included | ICD-10 + CPT/HCPCS populated |
| 5 | Disclaimer about variability and dispute rights | Template language recited + on PDF |
| 6 | Patient can request GFE; scheduled service auto-triggers | Consent capture on call |
| 7 | Delivered in language patient requests | 29 language support |
| 8 | Accessible (alternative formats on request) | SMS, email, paper mail options |
| 9 | Estimate retained for at least 6 years | Encrypted storage with retention policy |
| 10 | Dispute resolution process explained | Scripted explanation with contact info |
Every CallSphere financial counseling call satisfies all ten requirements through a combination of the voice conversation and the post-call document delivery. The auditable trail includes the call recording, the transcription, the generated PDF, and the delivery confirmation — all retained for the six-year regulatory window.
The three-business-day delivery window is the most commonly missed NSA requirement in CMS audits. CallSphere's workflow prevents this by generating the PDF GFE within 90 seconds of call end and delivering via SMS, email, or both. If the patient requests paper mail, a fulfillment task fires to the hospital's print-and-mail vendor with a 1-business-day SLA. The compliance attestation record logs the delivery method, timestamp, and confirmation — which is exactly what CMS auditors ask for.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for behavioral health in your browser — 60 seconds, no signup.
Hospital financial counseling splits into four workflows, each of which an AI voice agent handles differently.
Patient calls to schedule a service. The agent uses `get_services` to retrieve the CPT code and base charge, `get_patient_insurance` to determine whether the patient is uninsured or self-pay, and `get_providers` to identify expected co-providers (anesthesiology, radiology, pathology). The agent walks the patient through the expected charges, explains the estimate is an estimate (not a guarantee), recites the dispute rights disclaimer, and generates the PDF.
Patient calls with a bill in hand. The agent looks up the account, walks the itemized bill line by line, translates medical codes to plain-English descriptions, and explains insurance adjustments. This is where AI voice agents shine — they never lose patience explaining why the "CT abdomen with contrast" line is different from the "contrast agent" line, or why the deductible applied differently in January than in November.
For balances the patient cannot pay in full, the agent offers the hospital's standard payment plan options (typically 6, 12, or 24 months at 0% interest for amounts under $5,000). The agent captures the plan selection, calculates the monthly amount, confirms the payment method, and writes the plan into the revenue cycle system. A plan summary document is SMS'd to the patient.
Patients below 400% of federal poverty level typically qualify for charity care under the hospital's financial assistance policy (IRS 501(r) requirement). The agent screens eligibility, explains the application process, captures initial documentation via secure upload links, and creates a case for the financial counselor to review. The human counselor then only touches applications that are already partially complete, dramatically reducing their per-application time.
The CallSphere Revenue Cycle Maturity Model is an original five-stage framework that describes the progression of AI-enabled financial counseling from pilot to full automation. Most hospitals enter at Stage 1 and reach Stage 3 within 12-18 months.
| Stage | Name | Capabilities | Typical Hospital Outcome |
|---|---|---|---|
| 1 | Voice Triage | AI answers, classifies, routes to humans | 30% call deflection, 22% handle time reduction |
| 2 | GFE Automation | AI delivers NSA-compliant estimates end-to-end | 90%+ NSA compliance rate, 3x estimate delivery volume |
| 3 | Full Bill Explanation | AI handles bill questions and payment plans | 65%+ call automation, 18% collections uplift |
| 4 | Assistance Integration | AI pre-screens and collects charity care docs | 40% increase in FA application throughput |
| 5 | Proactive Outreach | AI initiates outbound estimates, reminders, and plan check-ins | 12-15% bad-debt reduction |
The stages are not sequential in implementation (most hospitals deploy Stages 1 and 2 simultaneously), but they are sequential in operational maturity — you do not run Stage 5 outbound reliably until Stage 2 inbound is stable.
The financial counseling agent sits on top of the hospital's revenue cycle system (Epic Resolute, Cerner Patient Accounting, Meditech MAGIC) and pulls real-time account data through ADT and billing interfaces. The architecture separates the conversational layer (CallSphere voice agent) from the pricing engine (hospital chargemaster), from the document generator (PDF renderer + template library), from the compliance logger (audit trail).
``` +------------------+ | Inbound call | +--------+---------+ v +------------------+ +------------------+ | CallSphere Voice |<------>| OpenAI gpt-4o- | | (gpt-4o-realtime)| | realtime 2025-06 | +--------+---------+ +------------------+ | | Function calls (14 tools) v +------------------+ | Hospital RCM API | | - get_services | | - lookup_patient| | - get_insurance | +--------+---------+ | v +------------------+ | GFE PDF Generator| | + SMS/email | | + Audit Log | +------------------+ ```
The 14 function-calling tools include `lookup_patient`, `lookup_patient_by_phone`, `create_new_patient`, `get_patient_insurance`, `get_services` (with CPT/CDT codes), `get_providers`, and `get_office_hours`. These tools let the agent pull real-time chargemaster and insurance data so the estimate reflects the patient's actual coverage, not a generic list price.
CallSphere's post-call analytics generate five signals per call: sentiment score, lead/collection probability score (0-100), intent classification, satisfaction rating (1-5), and escalation flag. The collection probability score is particularly valuable for revenue cycle leadership — it predicts the likelihood the patient will pay within 60 days based on tone, commitment language, and payment method capture. Patients scoring below 40 get routed to a collection specialist for follow-up; patients scoring above 70 typically pay without further intervention.
| Capability | Human-Only | Generic IVR | CallSphere AI Voice |
|---|---|---|---|
| 24/7 availability | No | Yes | Yes |
| GFE delivery window compliance | 76% | 34% | 94% |
| Bill explanation handling | Yes | No | Yes |
| Payment plan setup | Yes | Limited | Yes |
| Language support | Limited | 2-3 | 29 |
| Cost per call | $7.80 | $0.45 | $2.65 |
| Avg queue time | 18 min | 0 min | 0 min |
| Abandonment rate | 34% | 51% | 3% |
| NSA compliance audit pass rate | Variable | N/A | 94% |
See our platform comparisons for more context on voice agent vendor selection: CallSphere vs Bland AI, CallSphere vs Retell AI, CallSphere vs Synthflow.
Financial counseling AI deployments have the cleanest ROI story in healthcare AI. The math is deterministic because every variable is measurable from existing revenue cycle reports.
For a 400-bed hospital with $480M gross revenue and 8% self-pay mix:
Still reading? Stop comparing — try CallSphere live.
See the behavioral health AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Beyond the collections lift, hospitals see HRSA 340B reporting efficiency gains, lower complaint rates (AHA 2025 data shows 41% reduction in billing-related patient complaints post-deployment), and measurable reductions in patient-provider dispute filings under NSA. McKinsey's 2025 Healthcare Operations survey identified AI-enabled financial counseling as having the highest 12-month ROI of any hospital administrative AI use case.
See our pricing and features pages for deployment scoping, or contact sales to model the ROI for your specific revenue profile.
Even well-designed financial counseling automation hits edge cases that require human judgment. Building a production-grade program means knowing which edge cases to automate, which to escalate, and which to instrument for continuous improvement.
Patients occasionally call disputing a bill they consider a surprise under NSA. The agent must recognize the pattern ("I didn't expect this bill" / "they said this was covered" / "I was told it would be free") and route to the hospital's NSA dispute resolution contact. The agent does not attempt to resolve the dispute on the call — that is a legal process with a 30-day clock under 45 CFR § 149.620. The correct behavior is to open a formal dispute ticket, provide the patient with the federal dispute process information, and escalate to a human financial counselor for case management.
IRS 501(r) requires nonprofit hospitals to maintain a written financial assistance policy (FAP) and screen every self-pay patient for eligibility. The agent pre-screens against the FAP thresholds (typically 200-400% of federal poverty level for full assistance, sliding scale above), collects preliminary income attestation, and triggers the formal application process. HFMA data shows that hospitals deploying AI pre-screening see a 47% increase in FAP applications completed, because the friction of the paper-form process was previously deterring eligible patients from applying at all.
When a patient mentions bankruptcy, active litigation, or legal guardianship, the agent immediately escalates to a specialized team. The Fair Debt Collection Practices Act and state-level medical debt laws impose specific restrictions on collections activity for patients in bankruptcy or under legal protection, and violations create regulatory exposure. The agent's role is to recognize the signal and route, not to parse the legal situation.
Medicare Secondary Payer (MSP) questionnaires are required for every Medicare beneficiary encounter and are a frequent source of billing confusion. The agent walks through the MSP questionnaire in plain language, captures responses, and writes them to the patient's account. CMS's MSP enforcement actions in 2025 totaled $1.8B in recoveries, making accurate MSP capture a revenue-integrity priority. AI voice agents produce substantially higher MSP completion rates than paper questionnaires because they can clarify questions in real time.
Yes. The No Surprises Act does not specify the delivery mechanism — it specifies content, timing, and accessibility requirements. 45 CFR § 149.610 is silent on whether a human or automated system delivers the GFE, provided all requirements (written document, three-day window, language access, dispute rights disclosure) are met. CMS's 2024 NSA Implementation FAQ Update #7 explicitly contemplated voice-automated delivery.
The No Surprises Act already contemplates estimate variability — the actual bill can be up to $400 higher than the estimate before the patient has dispute rights. CallSphere's GFE generation pulls from the hospital's chargemaster in real time, so the estimate reflects the same pricing a human counselor would produce. Systematic errors are caught by the post-call QA review and corrected upstream in the chargemaster or logic.
The AI agent can explain the prior authorization process, check whether a specific service requires PA under the patient's plan, and initiate the PA request via the hospital's existing workflow. Actual clinical appeal arguments remain with human staff. The agent handles roughly 70% of inbound PA-related questions without escalation.
The agent handles routine financial conversations. For complex situations — disputed bills, divorce-related custody of medical expenses, legal guardianship — the agent recognizes the complexity signal and transfers to a human financial counselor with a summary of what was discussed. The post-call sentiment score and escalation flag surface these automatically.
Yes. The NSA applies to any facility that provides scheduled services to uninsured or self-pay patients. CallSphere deployments include hospital systems, ambulatory surgery centers, imaging centers, and physician group practices. The workflows are the same; the chargemaster integration varies by EHR.
Stage the rollout. Start with Stage 1 (voice triage) to offload routine routing, then add Stage 2 (GFE automation). Human counselors shift to complex cases, charity care applications, and payer escalations. Most hospitals report higher job satisfaction among counselors post-deployment because they spend less time on repetitive calls and more on complex patient advocacy.
Yes, through PCI-DSS compliant payment processing. The card capture happens in a separate secure subsession that is excluded from call recording. CallSphere integrates with major hospital payment processors (InstaMed, Change Healthcare, Waystar) for the actual transaction while the voice agent orchestrates the user experience.
CallSphere supports native dialogue in 29 languages including Spanish, Mandarin, Vietnamese, Tagalog, Arabic, and Russian. NSA language access requirements are fully met — the agent delivers the GFE, explains dispute rights, and handles payment setup in the patient's preferred language without handoff to a translator. Our healthcare AI overview covers the multilingual architecture in detail.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Graphiti is the open-source temporal knowledge graph for AI agents in 2026. Learn how bi-temporal memory beats vector RAG for voice agents and long-running LLMs.
How to design a multi-agent system using MCP for tools and A2A for cross-vendor coordination, with a CallSphere voice agent as a participating node.
Self-correction is now a property of the model, not the framework. What that means for production agent reliability, voice/chat fallbacks, and CallSphere.
Offline evals catch regressions before deploy on a fixed dataset. Online evals catch real-world drift on live traffic. You need both — here is how we run them.
Build a working voice agent with the OpenAI Realtime API + Agents SDK, then bolt on an eval pipeline that catches barge-in failures, hallucinated grounding, and latency regressions.
The full metric set for evaluating production voice agents — STT word error rate, end-to-end latency budgets, RAG grounding, prosody, and the metrics that actually correlate with retention.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI