We’re putting $5.5M behind research and development of smaller, specialized AI models for real-world deployment — Read our manifesto →
    All posts

    [Case Study] 14,583 Calls in One Day: Houston Methodist's Voice AI Surge Test

    How a conversational hotline answered every vaccine call on the first ring, automated 91% of intents, and protected existing contact-center capacity.

    Conscious Engines

    Industry: Healthcare
    Organization: Houston Methodist
    Use case: Voice automation for vaccine information, eligibility, and scheduling
    Evidence basis: Houston Methodist Center for Innovation case study
    Disclosure: This is an independent analysis by Conscious Engines. The use case came from the January 2021 vaccine rollout, so it is evidence of surge handling rather than a current general patient-access benchmark.

    1. Outcome at a Glance

    Houston Methodist's vaccine voice assistant handled more than 200,000 calls in its first month, including 14,583 calls in one day and as many as 3,500 calls in one hour. The health system reported a 91% automation rate across patient intents.

    Key Outcomes

    More than 200,000

    First-month calls

    January 1 to February 1, 2021 reported in the cited case.

    More than 9,000

    Average weekday calls

    Surge volume reported in the cited case.

    4,600

    Average weekend calls

    Surge volume reported in the cited case.

    14,583 calls

    Peak single day

    Elasticity test reported in the cited case.

    3,500 calls

    Peak hour

    Concurrency test reported in the cited case.

    MeasureReported resultEvidence interpretation
    First-month callsMore than 200,000January 1 to February 1, 2021
    Average weekday callsMore than 9,000Surge volume
    Average weekend calls4,600Surge volume
    Peak single day14,583 callsElasticity test
    Peak hour3,500 callsConcurrency test
    Automation rate91%Across reported patient intents
    Calls answered100% on first ring, 24/7No reported abandonment
    Eligibility and scheduling path75% of callersChecked eligibility and scheduled or joined follow-up line
    FAQ path9% of callersAccessed eligibility or vaccine information
    Vaccines deliveredMore than 4,000 per dayHealth-system output, not caused solely by voice AI

    The voice assistant protected existing operators and nurses from an expected 300% to 400% increase in call volume. Houston Methodist also avoided temporary staffing and additional telephony seats, although it did not publish a dollar saving.

    2. The Operational Problem

    Before vaccine availability, the health system expected a sudden wave of calls from patients and the public. Hiring enough people was too slow and expensive. Outsourcing threatened consistency and control of the patient experience.

    The use case combined information and action. Callers wanted to know whether they were eligible, understand safety and efficacy, schedule a dose, or register interest for a later phase. Some needed a live agent or nurse.

    The demand pattern was extreme and uncertain. A traditional IVR menu could route calls but would not understand varied questions or guide a caller through eligibility and scheduling.

    3. What Was Built

    Houston Methodist and Syllable created a dedicated phone-based vaccine system using a conversational voice assistant.

    System at a Glance

    Telephony entry

    Route vaccine calls away from normal hospital operators.

    Speech recognition

    Understand caller questions and selections.

    Eligibility logic

    Apply the current phase rules.

    Knowledge response

    Answer controlled vaccine FAQs.

    Scheduling workflow

    Start self-service appointment or future-contact registration.

    Human escalation

    Connect callers to an agent or nurse when needed.

    LayerFunction
    Telephony entryRoute vaccine calls away from normal hospital operators
    Speech recognitionUnderstand caller questions and selections
    Eligibility logicApply the current phase rules
    Knowledge responseAnswer controlled vaccine FAQs
    Scheduling workflowStart self-service appointment or future-contact registration
    Human escalationConnect callers to an agent or nurse when needed
    Elastic infrastructureAbsorb large hourly and daily volume changes

    The design is task specific. Eligibility should be deterministic and versioned as policy changes. The language model or intent system handles varied wording, while the workflow controls the actual action.

    This separation remains relevant. A modern voice agent should not infer eligibility from general knowledge. It should call a governed rule service and confirm the input fields.

    4. How It Reached Production

    Houston Methodist created one clear front door for vaccine demand.

    Isolate surge traffic. A dedicated hotline and modified operator greeting protected ordinary patient calls.

    Design for elastic peaks. Average volume would have hidden a 3,500-call hour. Capacity testing must use worst-case concurrency.

    Automate bounded intents. Eligibility, FAQs, scheduling, and future notification had clear outcomes.

    Retain clinical escalation. Vulnerable patients and complex situations could reach agents or nurses.

    Update policy rapidly. Vaccine phases changed frequently. The system needed versioned rules and content owners.

    For a current deployment, additional controls should cover identity, consent, accessibility, multilingual performance, emergency language, security, and transaction audit.

    5. What Healthcare Leaders Should Take Away

    This case remains powerful because it tested healthcare voice AI under real surge conditions. The system answered every call on the first ring while handling a peak of 14,583 calls in a day.

    The architecture can support appointment access, referrals, results routing, medication reminders, and administrative triage. Domain speech models, TTS, workflow rules, private RAG, system integrations, and human escalation should operate as one service.

    The primary metric is cost per correctly completed call objective, supported by first-call resolution, abandonment, transfer, safety escalation, and rapid repeat contact. Houston Methodist proved elasticity and automation. A new buyer should separately prove long-term economics and patient satisfaction.

    Sources