Clearwave

Clearwave sells eight agents across the front desk (voice, scheduling, eligibility, clinical intake, pre registration, check in, payments and outreach), coordinated by a workflow agent included in every plan. The voice agent is the module a model makes possible. It answers inbound calls in conversational language, in 33 languages, and books, confirms, cancels and reschedules against the practice's own scheduling rules; the company's facts page puts it at more than 50,000 calls a month. That capability does not exist without speech and language models, and it is sold as its own product.

Most of the rest stands without a model. Eligibility checks run as transactions against more than 900 payer networks, and registration, check in and payments are self service workflows the company sold as kiosks and digital check in before the agent relaunch. Scheduling is described as running on AI and rules based logic, with AI provider schedule optimization listed as a feature.

The agents act alone on routine work, and the company describes where they stop. The voice agent answers questions about hours, locations and directions, and books, confirms, cancels and reschedules using the same provider rules the practice sets for online scheduling. Clinical, prescription refill, billing and administrative requests are not completed by the agent: it writes them as tasks into the practice management system or EHR for staff. The voice FAQ says complex or exception based calls get a warm handoff to staff, and the facts page says a workflow agent orchestrates hand offs between agents and escalates exceptions to staff.

What is missing is the threshold. Nothing defines what counts as complex or an exception, whether a caller can simply ask for a person, how a caller describing urgent symptoms is handled, or what happens after a wrong booking. The company's coverage claims set the scope this has to hold across: it says the voice agent ensures every call is answered and handles 90 percent of incoming calls.

Eight agents and an orchestrator are described, and nothing in them is identified. The facts page describes eight task agents coordinated by a workflow agent that orchestrates hand offs and escalates exceptions, scheduling that matches patients to the correct provider using AI and rules based logic, AI provider schedule optimization, dynamic questioning in check in and intake, and a voice agent working in 33 languages with language matching.

No model, speech recognition or language provider is named, training data is not characterized, and no versioning or release practice is published. A buyer cannot tell from the site which agents contain a learned model and which are the earlier registration and eligibility workflows under a new name.

Payment partners are named; the model providers are not. The voice agent needs speech recognition, language understanding and speech synthesis, and none of those suppliers is identified. Two other data paths are partly visible. Eligibility runs against more than 900 payer networks, with no clearinghouse named. Card payments, card on file and surcharging run through preferred payment vendors, and the partnerships page names Rectangle Health and InstaMed as payment partners, alongside Rivet for patient estimates and Valer for prior authorizations and referrals.

No subprocessor list is published, and the privacy policy covers the website only.

The company publishes operating figures and names customers, with no method behind any number. Its facts page reports more than 5,000 locations and 11,000 providers, 2.5 million bookings, 150 million check ins and 2 billion eligibility checks a year, and more than $1 billion in patient payments captured to date. Those are volumes, not evidence of benefit.

The outcome figures are vendor reported percentages: front desk workload down by up to 89 percent, point of service collections up 154 percent, wait times down by up to 90 percent, claim denial prevention above 95 percent, and about 96 percent patient adoption of self check in in typical deployments. The voice agent is said to resolve calls 63 percent faster than a human agent's four minute average handle time and to handle 90 percent of incoming calls; neither figure states which calls were counted, over what period, or what counts as handled. Named customer stories on the homepage (Spire Practices, Sheltering Arms, CompEye, Marietta Eye, Holyoke Medical Center, Utah Cancer Specialists) carry further percentages.

No peer reviewed study, independent evaluation or error rate for any agent appears on the company's pages. The figure a practice most needs for the voice agent, how often it books the wrong slot or routes a request to the wrong place, is not published.

General assurances only. The company says it maintains HIPAA and SOC 2 compliance and describes registration as private and secure. Nothing on its homepage, facts, pricing, voice, platform or privacy pages states how long anything is kept, whether data is de identified, or whether customer data trains the agents.

Those gaps weigh heavily here. The platform collects scans of insurance cards and driver's licenses, clinical intake answers, card data and the audio of more than 50,000 calls a month across 33 languages. Call audio is the material a voice agent improves on, and it cannot be de identified the way a form field can. Kiosks and tablets run patient sessions on shared devices in waiting rooms, and nothing says what persists on one between patients. Nothing describes how the voice agent's answers are constrained, either.

Compliance is claimed and nothing supports it. The facts page, linked from the site footer, says Clearwave maintains SOC 2 and HIPAA compliance. No business associate statement or agreement appears on the company's pages, and the only privacy policy (last modified 1 May 2026) covers the website and says it does not apply to information submitted through the company's apps or patient portal.

That leaves the product's handling undocumented for data that includes scans of insurance cards and driver's licenses, demographics, consents, clinical intake answers, card payments and call audio, and leaves open whether subcontractors such as the payment vendors or a speech provider would be bound.

The facts page, linked from the site footer, states that Clearwave maintains SOC 2 compliance. That is a recognized attestation named in the company's own material. No report type, audit period or trust portal is published, so the report itself is not visible.

Two gaps sit beside it. Nothing names a payment card industry attestation, although the payments agent takes card payments at kiosks, on mobile and during pre registration, and card on file capture runs through preferred payment vendors. And unattended kiosks and tablets in waiting rooms need physical controls, such as card reader protection, session timeout and handling of lost devices, that nothing describes.

The product is scoped as front office work and makes no device claim: scheduling, call handling, registration, eligibility, payments and reminders. The clinical intake agent collects a patient's history before the visit rather than interpreting it. Where the voice pages say the agent triages requests, they mean routing: prescription refills and clinical questions are written as tasks for the practice's clinical team, not assessed by the agent.

Other regimes touch the product. Card payments bring payment card industry requirements, and the company describes its surcharging at check in as compliant for clients on its preferred payment vendors, without saying which state rules or card network rules that covers.

Nothing on the company's pages describes how the agents' behavior is evaluated or governed: no fairness statement, no accuracy by language, no testing across caller groups. The voice agent is marketed on reach and recognition: 33 languages, matching each caller's language preference, and meeting callers' needs accurately.

That is where the risk sits. Published research found speech recognition errors roughly twice as frequent for Black speakers as for white speakers across five commercial systems (Koenecke and colleagues, PNAS, 2020), and a caller the agent fails to understand can hang up without leaving a complaint or a record. The same question applies to accents, older callers and each of the 33 languages, for none of which is performance published. Self service check in raises a related access question: the roughly 96 percent adoption figure counts patients who completed it, not who needed help. Nor is it said how calls the agent could not complete are counted.

Nothing on the company's pages says what happens when an agent is wrong. The eligibility agent's result decides what a patient is told they owe and whether a claim is later denied. The payments agent collects copays, balances and surcharges on an automated determination of what is owed. The scheduling and voice agents book, cancel and reschedule without staff. No error rate, accuracy commitment, refund or dispute route, indemnity or remedy is published for any of them, and the pricing page ties price to practice size, patient volume and providers rather than to results.

Patients can change their own bookings through Appointment Manager, and exceptions reach staff, but nothing lets a patient or practice see a wrong eligibility result or an overcollection unless it surfaces later.

Integration is broad and described as two way. The integrations page says the agents write directly back into the practice management system and EHR, and that Clearwave holds formal API license or partnership agreements with named record companies including athenahealth, Epic, MEDITECH, CareTracker, Compulink, Modernizing Medicine and Unlimited Systems. The homepage links dedicated integration pages for athenaOne, ModMed, Veradigm, MEDITECH, Unlimited Systems, NextGen and eClinicalWorks. Bookings, registration data and payments post into the practice's systems, and the voice agent writes refill and records requests there as tasks. Since September 2026 the scheduling agent also takes bookings through Reserve with Google from Search and Maps. Eligibility runs against more than 900 payer networks.

The partnership claims are the vendor's own. The athenahealth Marketplace listing the integrations page cites is the one outside place a connection can be confirmed, and no public API documentation is published for customers building their own workflows.

Nothing on the company's homepage, facts, pricing, voice, platform or privacy pages says where the platform is hosted, in which region data rests, or how one customer's data is separated from another's. The privacy policy covers the website only.

Part of the deployment is physical. Kiosks and tablets in waiting rooms run patient sessions containing identity documents, clinical answers and card data on devices in public space, and nothing describes what is stored on them or how a lost device is handled. The voice agent adds telephony and speech processing whose location is not stated either. No uptime commitment is published, and a practice that has reduced front desk staff on the strength of the platform has little manual fallback.

The pricing page linked from the footer explains how the product is sold and gives no price. Pricing is a custom quote shaped by practice size, patient volume, provider count and other requirements, with an ROI calculator keyed to the number of providers. The facts page adds that clients select the agents they need and that the workflow agent coordinating them is included with every plan, so the product is modular by agent.

What a buyer cannot establish before contact is the unit of charge: per provider, per location, per visit, per call, or a share of collections. The last matters because the payments agent collects copays, past due balances and card surcharges, and the company describes surcharging through preferred payment vendors. Hardware terms for kiosks and tablets are not stated.

Built for specialty practices and multi site groups, with hospitals and health systems as a second segment. The site carries dedicated pages for ophthalmology, orthopedics, oncology, dermatology, ENT and allergy, cardiology, gastroenterology, neurology, obstetrics and gynecology, internal and family medicine, and federally qualified health centers. The company reports more than 5,000 locations and 11,000 providers, and the homepage names customer stories by specialty, from orthopedic groups of more than 100 providers to a 59 provider hospital and ambulatory system.

Channels cover the whole front desk: inbound calls in 33 languages, online booking including Google Search and Maps, pre visit links, mobile, tablets and waiting room kiosks. Eligibility reaches more than 900 payer networks. Validation sits in the customer stories, which are the vendor's own; nothing independent confirms performance outside the specialties they name.