By Sagar Shankaran, Founder of CallSphere
How ENT (otolaryngology) practices use AI voice agents to handle hearing aid trial follow-ups, allergy surge capacity, and sleep study (PSG) scheduling without adding staff.
Key takeaways
ENT practices combine three very different workflows under one phone number: high-acuity procedures (tonsillectomy, sinus surgery, sleep surgery), chronic longitudinal management (hearing aids, allergy, tinnitus), and seasonal surges (spring and fall allergy peaks can 3x inbound call volume for 6–8 weeks). Traditional staffing cannot elastically expand for allergy season, cannot run the structured 30/60/90-day hearing aid fitting follow-up cadence recommended by the American Academy of Audiology, and cannot triage a "ringing in my ear" call correctly at 8pm. An AI voice agent on OpenAI's `gpt-4o-realtime-preview-2025-06-03` model scales to arbitrary concurrent call volume, runs deterministic hearing aid follow-ups, and routes sleep study scheduling between polysomnography (PSG) and home sleep apnea testing (HSAT) based on AASM criteria.
According to the Vision Council / Hearing Industries Association 2024 MarkeTrak 2024 study, 28.8 million U.S. adults could benefit from hearing aids but only 19% have them, and 15–20% of those who do try hearing aids abandon them within the first 90 days — a number that drops to 6–8% when practices run structured follow-up at 30, 60, and 90 days. That is a voice-agent-sized problem. CallSphere's ENT deployment uses the healthcare agent's 14 tools (`lookup_patient`, `get_available_slots`, `schedule_appointment`, `get_patient_insurance`, `get_providers`, and others) plus the after-hours escalation ladder with its 7 agents, Twilio call+SMS fallback, and 120s per-agent timeout.
The ENT Call Routing Elasticity Model (CREM) is CallSphere's original framework for matching ENT call types to service tiers under variable load. It classifies every inbound call on three axes: urgency (emergent, urgent, routine), category (surgical, medical, audiology, sleep, allergy), and acuity score (0–10 from symptom capture). The matrix routes the call to one of five tiers — in-agent completion, async callback, same-day triage, immediate warm transfer, or 911/ED referral.
Spring allergy volumes surge to approximately 3.2x baseline per a 2023 AAO-HNS practice survey, while audiology call volume is relatively flat year-round. The CREM lets the practice set load-shedding rules: during allergy surge, route all allergy refill requests directly to the voice agent (which uses `lookup_patient` + `get_patient_insurance` + a formulary check), freeing human staff for surgical and sleep calls that need judgment.
| Tier | Call Type Example | Handling | Avg Call Duration |
|---|---|---|---|
| T0 — In-agent | Allergy refill, appt reschedule | 100% autonomous | 90 sec |
| T1 — Async callback | Hearing aid cleaning question | Agent captures, schedules callback | 60 sec |
| T2 — Same-day triage | "Sudden hearing loss" | Warm transfer to audiologist same day | 120 sec + transfer |
| T3 — Immediate transfer | Severe epistaxis, post-op bleeding | Warm transfer via 7-agent ladder | < 90 sec |
| T4 — 911/ED | Airway compromise, stridor | Explicit 911 instruction + hold on line | Call maintained |
| Season | Baseline Daily Calls | Peak Daily Calls | Staff Required (Human Only) | With Voice Agent |
|---|---|---|---|---|
| Winter | 180 | 220 | 3 FTE | 1 FTE + agent |
| Spring allergy | 180 | 580 | 9 FTE (impossible) | 1 FTE + agent |
| Summer | 180 | 240 | 3 FTE | 1 FTE + agent |
| Fall allergy | 180 | 510 | 8 FTE (impossible) | 1 FTE + agent |
American Academy of Audiology best practice is a structured 30/60/90-day follow-up for every hearing aid fitting, covering fit/comfort, acoustic satisfaction, program usage, and return-for-credit decision before the manufacturer return window closes (typically 45–60 days). Missing a follow-up in this window is a direct revenue loss: the patient returns the aids, the practice absorbs restocking fees, and the clinical relationship ends. MarkeTrak 2024 found practices with structured follow-up have 92–94% 90-day retention versus 78% without.
The voice agent runs three scheduled outbound calls — 30, 60, and 90 days post-fit — executing the exact same standardized questions each time so outcomes are comparable across patients. Each call writes a structured satisfaction payload to the EHR and flags any C-level concern (unable to hear in noise, feedback, discomfort) for the audiologist.
```typescript // CallSphere hearing aid follow-up state machine type HAFollowupWindow = "day_30" | "day_60" | "day_90";
interface HASatisfactionPayload { patientId: string; window: HAFollowupWindow; fitComfort: 1 | 2 | 3 | 4 | 5; soundQuality: 1 | 2 | 3 | 4 | 5; dailyWearHours: number; feedbackOccurring: boolean; programsUsed: string[]; rlikelihoodToKeep: 1 | 2 | 3 | 4 | 5; openConcerns: string; escalationNeeded: boolean; }
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
async function scheduleHAFollowup(patientId: string, fitDate: Date) { for (const offset of [30, 60, 90]) { await scheduler.enqueue({ patientId, callAt: addDays(fitDate, offset), script: `ha_followup_day_${offset}` }); } } ```
| Window | Core Questions | Escalation Trigger | Typical Outcome |
|---|---|---|---|
| Day 30 | Comfort, wear time, battery management | < 4 hr/day wear, any pain | In-person re-fit |
| Day 60 | Noise performance, program switching | Feedback ongoing, satisfaction < 3 | Re-program |
| Day 90 | Long-term satisfaction, return decision | Likelihood-to-keep < 3 | Audiologist call before return window |
Spring and fall allergy peaks reliably push ENT practices past staffing capacity for 6–8 weeks each season. The dominant call categories during surge are refill requests (antihistamine, intranasal steroid, leukotriene receptor antagonist), injection-schedule questions for patients on subcutaneous immunotherapy (SCIT), and symptom-severity escalations. An AI voice agent handles refills and schedule questions autonomously and routes symptom-severity cases to the appropriate tier.
The CDC estimates approximately 26% of U.S. adults and 19% of children have seasonal allergies. In a typical 10,000-patient ENT practice, that implies 2,000–3,000 allergy-active patients, of whom roughly 35% call at least once during peak season. The voice agent's capacity is effectively unbounded — 200+ concurrent calls on a single Twilio trunk — so surge does not translate to hold times.
| Call Reason | % of Allergy Calls | Voice Agent Handling |
|---|---|---|
| Refill request | 42% | `lookup_patient` + refill + `schedule_appointment` if > 1yr since visit |
| SCIT injection question | 18% | Confirm schedule, check reaction history |
| Symptom escalation | 22% | Acuity-scored, T1/T2/T3 routing |
| Appointment scheduling | 14% | `get_available_slots` + `schedule_appointment` |
| Billing / insurance | 4% | `get_patient_insurance` + routing |
The American Academy of Sleep Medicine (AASM) Clinical Practice Guideline for Diagnostic Testing for Adult OSA distinguishes between in-lab polysomnography (PSG) and home sleep apnea testing (HSAT) based on patient characteristics: HSAT is appropriate for uncomplicated adults with high pre-test probability of moderate-to-severe OSA; PSG is required for patients with significant comorbidities (CHF, COPD, neuromuscular disease), suspected non-OSA sleep disorders, or negative HSAT with persistent suspicion. A voice agent that captures STOP-BANG, Epworth, and comorbidity status during the scheduling call selects the correct test on the first try — avoiding the common failure mode of "patient did HSAT, was inconclusive, had to re-schedule PSG 6 weeks later."
An estimated 30 million U.S. adults have OSA per the American Academy of Sleep Medicine, but only 6 million are diagnosed. Each undiagnosed case carries ~$1,400/year in excess Medicare spend per CMS data. Sleep study throughput is the bottleneck; accurate test selection at scheduling time is the lever.
| Patient Profile | STOP-BANG | Comorbidities | Recommended Test | Insurance Pre-Auth |
|---|---|---|---|---|
| Adult 30–65, uncomplicated | >= 3 | None major | HSAT | Most plans no PA |
| Adult with CHF | Any | CHF EF < 45% | PSG | PA required |
| Adult with COPD | Any | FEV1 < 50% | PSG | PA required |
| Adult with neuromuscular | Any | ALS, MD, etc. | PSG | PA required |
| Pediatric (< 18) | n/a | Tonsillar hypertrophy | PSG | PA required |
| Post-treatment assessment | n/a | Treated OSA | HSAT or PSG | PA + medical necessity |
The agent pulls comorbidity codes via `lookup_patient`, runs STOP-BANG verbally, and uses `get_patient_insurance` to check PA requirements. It schedules via `get_available_slots` + `schedule_appointment` with the correct test type pre-selected.
Tinnitus and balance disorders make up roughly 9% of ENT ambulatory visits per AAO-HNS practice benchmark data, and they generate disproportionately high call volume because both conditions are chronic, symptom-fluctuating, and anxiety-provoking. A tinnitus patient typically calls 3–5 times per year between visits asking whether the symptom is worsening, whether a new sound indicates something serious, or whether a new supplement is appropriate. The voice agent handles education, symptom logging, and routing; it does not dispense clinical advice. Persistent unilateral tinnitus, pulsatile tinnitus, or tinnitus associated with sudden hearing loss all route to Tier 2 or Tier 3 per AAO-HNS Clinical Practice Guideline on Tinnitus (2014, updated 2020).
Balance complaints route based on BPPV screening questions (positional vs constant, duration, associated hearing loss). Acute vertigo with neurologic symptoms is a Tier 4 (911/ED) call per AAO-HNS guidance. Episodic BPPV-pattern vertigo routes to audiology or vestibular PT same or next day. The agent captures Dizziness Handicap Inventory (DHI) responses by voice when a longitudinal patient calls.
| Symptom | Tier | Agent Action |
|---|---|---|
| Bilateral tinnitus, stable | T0/T1 | Log, educate, schedule routine |
| New unilateral tinnitus | T2 | Same-day audiology evaluation |
| Pulsatile tinnitus | T2 | Urgent evaluation, imaging prep |
| BPPV-pattern positional vertigo | T1 | Schedule vestibular assessment |
| Vertigo + neuro symptoms (weakness, speech) | T4 | 911 instruction, maintain line |
| Chronic Meniere's flare | T2 | Same-day physician call |
ENT practices run a heavy post-operative call load — tonsillectomy Day-5 bleeding checks, sinus surgery debridement scheduling, and post-thyroidectomy voice monitoring. Tonsillectomy post-op bleeding is a well-defined risk window peaking around post-op Day 5–7 per AAP tonsillectomy guidelines. The voice agent runs proactive Day-3, Day-5, and Day-7 outbound check-ins for every pediatric and adult tonsillectomy patient, asking about pain control, hydration, fever, and any bleeding episodes. Any bleeding report — even small, self-limited — triggers an immediate physician call.
Similarly, post-FESS (functional endoscopic sinus surgery) patients get Day-2, Day-7, and Day-14 check-ins coordinating saline rinse compliance, debridement scheduling, and symptom monitoring. The AAO-HNS reports post-FESS follow-up compliance is the strongest predictor of surgical success; practices that systematize these calls see 18–22% fewer revision surgeries per a 2023 Otolaryngology–Head and Neck Surgery journal analysis.
Every call produces a structured outcome record: reason, tier, disposition, tools invoked, revenue attributed, QA flags. Post-call analytics aggregate these into weekly dashboards the practice administrator uses to (a) right-size staffing around real demand, (b) identify bottlenecks (e.g., sleep study scheduling is 14% of calls but 31% of avg duration), and (c) measure campaign impact. The same engine powers the pricing breakdown by tier and the features catalog.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
The after-hours escalation system handles the 8pm "sudden hearing loss" call with a 7-agent rotation, Twilio call+SMS ladder, and 120s per-agent timeout — the same plumbing described in the therapy practice guide and the AI voice agents in healthcare overview.
Pediatric ENT volume — tonsillectomy, adenoidectomy, and pressure equalization (PE) tube placement — concentrates heavily in the 2–8 age range and carries its own communication pattern. Parents of post-op pediatric patients have more questions, higher anxiety, and are more likely to call at non-business hours. The voice agent handles parent-facing scheduling, pre-op prep coordination, post-op check-ins, and symptom capture on the same tiered routing model, with warm transfer to the on-call for any bleeding, airway, or fever concerns post-tonsillectomy.
PE tube placement is the most common pediatric surgical procedure in the U.S., with roughly 667,000 performed annually per AAO-HNS data. Post-operative follow-up at 2 weeks and 6 weeks is standard; the voice agent schedules and reminds both. Tube extrusion and persistent otorrhea are common call reasons — routine, but requiring same-day assessment when persistent. The agent captures symptom duration, discharge characteristics, and fever, routing appropriately.
| Procedure | Follow-up Windows | Typical Symptom Calls | Tier |
|---|---|---|---|
| Tonsillectomy | Day 3, 5, 7, then 2-week visit | Pain, hydration, fever, bleeding | T2/T3 for bleeding |
| Adenoidectomy | Day 3, 2-week visit | Nasal congestion, fever | T1 typically |
| PE tubes | 2 weeks, 6 weeks, 6 months | Drainage, hearing, tube status | T1/T2 |
| Septoplasty (adolescent) | Week 1, Week 4 | Nasal breathing, crusting | T1 |
A typical 5-provider ENT practice with 18,000 active patients, mixed surgical/medical/audiology, sees the following Year 1 impact from a voice agent deployment: (1) $220,000–$380,000 in recovered revenue from audiology recall and hearing aid retention, (2) $120,000–$210,000 in sleep study throughput improvements (fewer mis-scheduled tests, shorter time-to-diagnosis), (3) 1.0–1.5 FTE of front-desk labor redirected from phone work to clinical support, (4) measurable reduction in allergy-season hold-time abandonment (from 22% to under 3%), (5) quality-score improvements that unlock commercial and Medicare quality bonuses. The monthly subscription typically lands in the low-to-mid four figures depending on call volume and integration complexity.
| Metric | Before Agent | After Agent | Delta |
|---|---|---|---|
| Inbound call abandonment | 18% | 2% | -16 pts |
| Hearing aid 90-day retention | 76% | 92% | +16 pts |
| Annual exam recall close rate | 41% | 84% | +43 pts |
| Sleep study mis-routing rate | 14% | 3% | -11 pts |
| Front-desk FTE | 4.0 | 2.5 | -1.5 FTE |
| Net Year 1 revenue recovered | — | $340k–$590k | positive |
Yes. Sudden sensorineural hearing loss (SSNHL) is a Tier 2 (same-day triage) or Tier 3 (immediate) call depending on duration and associated symptoms. The AAO-HNS Clinical Practice Guideline on SSNHL recommends evaluation within 14 days with steroids strongly considered in the first 2 weeks. The agent captures onset timing, unilateral vs bilateral, vertigo presence, and routes to same-day audiology if < 48 hours or immediate transfer if associated with facial weakness.
It runs STOP-BANG plus a comorbidity screen pulled from `lookup_patient`. Uncomplicated adults with STOP-BANG >= 3 and no major comorbidities route to HSAT; patients with CHF, significant COPD, neuromuscular disease, or pediatric age route to PSG. It checks `get_patient_insurance` for PA requirements before booking. This cuts mis-scheduled tests to near zero.
The agent handles SCIT schedule questions — confirming the current vial, dose, and next injection date — and routes any prior-reaction or acceleration question to a clinician. It does not modify the schedule; that's a clinical call.
Yes. Routine cleaning and reprogramming appointments are Tier 0 (in-agent). The agent books them via `get_available_slots` and `schedule_appointment` with the right appointment type code for the EHR.
200+ concurrent calls per Twilio trunk. Spring allergy surge of 3.2x baseline (per AAO-HNS 2023) is handled without hold-time degradation because the voice agent's concurrency ceiling is 10x+ typical peak load.
At fitting, the audiologist's EHR note triggers a webhook to CallSphere's scheduler, which enqueues three outbound calls at fit_date + 30, + 60, + 90 days. Each call writes a structured satisfaction payload to the EHR. Concerning responses flag the audiologist before the next business day.
English and Spanish are native on `gpt-4o-realtime-preview-2025-06-03`. Other languages can be added via custom deployment; coverage depends on STT/TTS quality for the target language.
The most common ENT EHRs — Epic, Athena, eClinicalWorks, Modernizing Medicine EMA — are supported out of the box via FHIR or proprietary APIs. Others are 2–4 weeks of connector work. See contact for integration scoping.
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