Clearwave

A mature self service registration business that repositioned itself as an agentic platform two weeks before this assessment, and the grade reflects the gap between those two descriptions.

The established product is workflow and integration rather than intelligence. Kiosks, tablets and pre visit links collect demographics, consents and clinical history; eligibility checks run against more than 900 payers through transaction interfaces; payments are captured at the point of service. That machinery produces the impressive operational numbers on this record and almost none of it requires a model. Real time eligibility verification is a transaction against a payer, not a prediction.

The August 2026 launch names eight agents and two of them are plainly model driven. The voice agent answers inbound calls and books, confirms, cancels and reschedules, which requires speech recognition, intent handling and dialogue management, and it is running at more than 50,000 calls a month, so it is in production rather than announced. Scheduling is described as driven by booking logic complex enough to handle specialty provider rules, and dynamic questioning in intake adapts to appointment service, provider and location.

What holds this at C is that most of the named agents describe existing capabilities in new language. An eligibility agent running payer checks and a check in agent completing registration in under two minutes are the same functions the company sold before the rebrand. Renaming a workflow an agent does not make it a model, and a reader comparing this record to others should understand that the strongest artificial intelligence claim here is the voice layer.

Graded C: genuine and substantial conversational automation, sitting on a business whose value is mostly self service workflow.

The autonomy claim is the strongest in this session and the oversight description is the thinnest, which is an uncomfortable combination to record.

The company is explicit that the point is automation rather than assistance, positioning the platform as automating the work rather than as software that helps staff work faster, and describing the agents as absorbing work that traditionally falls on administrative staff. The volumes bear it out: a voice agent handling more than 50,000 calls a month at roughly 90 seconds each is operating without a person on most of those calls, and check in completing in under two minutes without staff intervention is the stated design.

Nothing published describes the boundaries. No escalation criteria, no handoff mechanism to a human, no statement of whether a patient can reach a person on request, no coverage hours for fallback, and no error handling for the cases where an agent misunderstands.

Two failure modes deserve naming. A patient calling a specialty practice may describe a symptom that should not wait for the next available appointment, and a booking agent optimised to fill the calendar has no clinical logic behind it, unlike the triage products on neighbouring records which are built precisely for that judgement. And a patient struggling with an unattended kiosk in a waiting room has no obvious path to help if the design assumes no staff intervention.

The honest mitigation is that these are administrative decisions with recoverable consequences. A wrong appointment is rescheduled; a failed check in is completed by a receptionist. That keeps this from grading lower.

Graded C: high autonomy, low stakes per instance, no documented controls.

The agent framing describes eight capabilities and discloses nothing about how any of them work, which leaves a reader unable to tell which are models at all.

What is published is a functional list: voice answering calls and booking, scheduling filling the calendar, eligibility checking payers, clinical intake collecting data, pre registration capturing demographics and consents, check in completing in two minutes, payments collecting balances, communications. Each is described by outcome. None is described by mechanism, and the word agent is applied uniformly across capabilities that are plainly different in kind, since an eligibility check against a payer transaction interface and a conversational voice agent have almost nothing technically in common.

That uniform labelling is itself the transparency problem on this record. A buyer reading the launch material cannot determine which of the eight involve a model, which are deterministic workflows renamed, or what changed technically at the August 2026 launch versus what was rebranded. Given the company sold most of these functions before the rebrand, that distinction is exactly what a diligence process needs.

Below the labels there is nothing at all: no architecture, no training data, no accuracy or error rates, no evaluation method, no versioning, no update cadence, and no named provider for speech recognition, language understanding or synthesis in the voice agent.

One useful technical detail does appear. Intake workflows are described as presented dynamically based on appointment service, provider and location, and dynamic questioning is named as an agent behaviour, which tells a buyer the questioning adapts rather than being a static form.

Graded D.

No upstream dependency is named for any component, and on a platform launched two weeks ago as an agentic workforce that is the disclosure a buyer most needs.

The voice agent is the conspicuous case. Answering more than 50,000 calls a month requires speech recognition, language understanding, dialogue management, speech synthesis and telephony, and a company of this profile is very unlikely to have built all of that in house in the period before an August 2026 launch. Nothing names a provider for any layer, and each unnamed provider is a party handling patient voice audio with its own retention and training terms.

Two further dependency categories are implied by the product and never enumerated. Eligibility across more than 900 payers runs through clearinghouse or transaction network infrastructure that the company does not operate itself, and that path carries protected information to a third party. Payment processing likewise requires an acquirer, gateway and card network relationships, none named, and payment card data flows through all of them.

So this record has three distinct third party data paths, for voice, eligibility and payments, and identifies none of them. For a customer conducting diligence, enumerating subprocessors is a basic requirement and there is no published starting point.

One piece of provenance is available and it is organisational rather than technical: the platform grew from a check in kiosk business over roughly a decade, so the registration and payments components are long standing in house products rather than acquired ones.

Graded D as an absence of disclosure rather than evidence of a problem.

Two very different bodies of evidence sit on this record and they should not be read as one.

The registration and payments evidence is mature and specific. Ninety nine percent patient adoption of registration, check in times reduced by 90 percent, collection rates of 96 percent during check in, point of service collections up between 85 and 154 percent, eligibility saving staff more than 500 hours a year, and cash flow accelerated by 65 percent. Named customers give operational detail rather than testimonial: an orthopaedic institute describes operating through a 50 percent reduction in registration staff, and a health system revenue cycle leader describes the impact directly. Adoption figures are the most credible of these, because patient adoption of a kiosk is directly observable and hard to inflate.

The agentic evidence is a two week old launch. The headline figures, 63 percent faster than human agents, an 89 percent drop in staff workloads and 86 percent less training time, carry no method, baseline, sample or comparison group. Faster than human agents at what task, measured how, against which staff, is unstated. The volume claims are more useful because they are counts rather than comparisons: more than 50,000 voice calls monthly at roughly 90 seconds average, and 150 million patient visits a year across check in and pre registration.

No peer reviewed work and no independent evaluation of any kind was located, and no error rate is published for any agent. For a voice agent booking appointments, the mis booking rate is the number a practice would want and it does not exist publicly.

Graded C: strong operational evidence for the mature product, promotional figures for the new one.

Nothing published addresses retention, secondary use or model training for any of the data this platform collects, and the collection surface is the widest on any record in this session.

What passes through it: insurance card and government identification images, demographics, consents, clinical histories and screener responses, payment card details, and voice recordings from more than 50,000 calls a month. Identification document images deserve particular attention because they are not clinical data and carry identity theft exposure that outlives any clinical relevance, and nothing states how long they are kept after eligibility is confirmed.

The voice agent raises the training question most directly. Fifty thousand calls a month is a substantial corpus of patient speech, it is exactly the material needed to improve a healthcare voice agent, and audio cannot be de identified the way a form field can. Whether recordings are retained, whether they train models, and whether learning happens per practice or across the customer base is unstated in either direction.

The unattended device surface is a stewardship question as much as a security one. Kiosks and tablets in waiting rooms hold session data on hardware in a public space, and nothing describes what persists on the device between patients.

One aspect of the design works in the vendor's favour. Passwordless patient led registration means the patient enters their own data rather than staff transcribing it, which reduces internal exposure and improves accuracy at the same time. That is a genuinely good property and it says nothing about what happens to the data afterwards.

Graded D on the absence.

No published position was located. No security or trust page, no certification claim, no protected data handling description, no business associate statement and no retention position were found on the vendor's site or in any third party material.

The absence is harder to overlook here than on most records because of the sheer breadth of sensitive data this product handles at a single touchpoint. Patients upload images of insurance cards and government identification. They enter demographics, complete consents, and provide clinical histories and screener responses before arrival. They make card payments. The voice agent records calls in which patients discuss appointments and, inevitably, symptoms. That is protected health information, government identification documents and payment card data collected through one interface, and the payment card element brings its own separate industry compliance regime that is also unmentioned.

The device deployment adds a physical dimension absent elsewhere in this index. Kiosks and tablets sit in public waiting rooms, unattended, running a session in which a patient enters clinical and financial information. Screen privacy, session timeout, data at rest on the device and what happens when a patient walks away mid registration are all real controls, and the company markets registration as private and secure without describing any of them.

Scale makes the silence conspicuous. A vendor describing 150 million patient visits a year has satisfied a great many procurement reviews and evidently holds the artefacts. None is published.

Graded D on published evidence, stated as an absence of disclosure rather than a judgement about the engineering.

No published security posture was located. No trust centre, no service organisation control claim, no HITRUST claim, no payment card industry attestation, no penetration testing statement, no vulnerability disclosure policy and no incident notification commitment were found.

The payment card omission is the most surprising. A platform collecting card payments through kiosks, mobile devices and a voice agent is unambiguously in scope for the payment card industry data security standard, that standard requires validated compliance rather than self assertion, and any practice accepting cards through this platform inherits scope from it. A vendor at this transaction volume certainly holds an attestation. It is not published, and unlike a service organisation control report there is no reason of confidentiality that would prevent stating that one exists.

The physical device estate compounds the gap. Unattended kiosks in public waiting rooms are a well understood attack surface, covering tampering, skimming on card readers, session persistence between users and physical access to the device itself. Card readers in unattended terminals have their own specific requirements under the same payment standard. None of this is addressed.

Scale makes the silence conspicuous rather than understandable. A vendor describing 150 million patient visits a year and integrations with practice management systems has passed a very large number of security reviews and holds the artefacts to prove it.

Graded D on published evidence. Of every gap on this record, this is the one most likely to be a publishing decision rather than a capability gap, and the cheapest to close.

No clearance is claimed and none is required. Scheduling, registration, eligibility verification, intake collection and payment processing carry no clinical claim, and the clinical intake agent gathers a patient's own history rather than interpreting it. The device question does not arise.

The regulatory exposure sits in three other places and none is addressed publicly.

Payments is the largest and the most concrete. A platform collecting card payments at kiosks, on mobile devices and over the phone is inside the payment card industry compliance regime, which is a genuine certification standard with its own audit requirements, and it is entirely unmentioned. The company also publishes commentary on credit card surcharging and describes the payments agent collecting surcharge fees, and surcharging is regulated at state level with several states restricting or prohibiting it and card network rules imposing disclosure requirements. A vendor automating surcharge collection across a multi state customer base is automating a regulated practice.

Outbound and inbound call handling engages telephone consumer protection rules for confirmations and reminders, and several states now require disclosure when a caller is interacting with an artificial agent rather than a person. Nothing describes whether the voice agent identifies itself.

Consent capture is the third. The pre registration agent collects consents, which are legal instruments, and nothing describes how validity is established when an agent rather than a person presents them.

Graded C: correctly outside device regulation, silent on three regimes that plainly apply.

Nothing was located. No model card, no training data description, no accuracy figures for any agent, no subgroup analysis and no bias statement.

The exposure here is about access to the front door, and it has a specific and well understood shape.

Voice recognition performance varies with accent, dialect, speech rate, age and speech impairment. A voice agent handling more than 50,000 calls a month as the primary route into a practice will therefore serve some callers materially worse than others, and the consequence is not a poor experience but a failure to obtain care: a patient who cannot get the agent to understand them, and who has no described path to a human, does not book. The company's stated benefit of no hold times and no missed calls is measured on calls the agent handles, not on callers who give up.

The self service design compounds it. Passwordless patient led registration on a kiosk assumes literacy, language fluency, digital confidence, dexterity and vision. The 99 percent adoption figure is a strong result and it is a measure of how many patients completed the process, not of who found it hard or who needed help that the staffing model no longer provides. A product marketed on operating with fewer registration staff has removed the fallback that older and less digitally confident patients previously relied on.

Nothing published addresses language coverage for either the voice agent or the registration interface, and nothing describes accessibility conformance.

Graded D on the absence, with the mechanism named because the adoption figure could easily be read as evidence that it does not exist.

Nothing published addresses responsibility for an automated outcome, and this platform generates the widest range of consequential errors of any record in this session.

Eligibility is the sharpest. A verification returning the wrong coverage status leads either to a patient being turned away or treated as uninsured, or to a claim denied after the fact and a bill the patient did not expect. The company markets reduced claim rejections and accurate collections at check in, so customers are relying on the output, and nothing describes accuracy commitments or who bears the loss when a check is wrong.

Payments is the second. A payments agent collecting copays, past due balances and surcharges is taking money from patients based on an automated determination of what they owe. An overcollection is a refund process at best and a complaint at worst, and surcharge collection is regulated at state level. Nothing describes reconciliation, dispute handling or error correction.

Scheduling and voice produce a third category. An agent that mis books, cancels the wrong appointment or fails to understand a caller creates delay, and at more than 50,000 calls a month a small error rate is a large absolute number.

No indemnity, limitation, performance warranty, service level or recourse route was located for any of it, and no error rate is published for any agent.

The stakes per instance are lower than on the triage records, since these are recoverable administrative and financial errors rather than clinical ones. The volume is far higher. Graded D.

Genuinely deep where it counts for this product, and narrower in reach than the enterprise records on adjacent pages.

The integration is bidirectional and operationally real. Registration and check in data flows into practice management and record systems, described as seamless handoffs, so a patient completing intake on a kiosk updates the system of record rather than producing a document someone rekeys. Scheduling writes into the practice's calendar with booking logic that respects complex provider rules, which requires live availability rather than a request queue. Eligibility runs real time checks against more than 900 payers and returns coverage status before the visit, which is a payer transaction network rather than a record system integration and is a substantial interoperability asset in its own right. Payments post against balances.

NextGen is named specifically with a joint integration page and marketed as a real time integration, which is more concrete than a marketplace logo.

What holds this at B is the breadth of named systems rather than the depth of any one. Kyruus Health on the adjacent record names four record systems covering effectively the whole United States market and publishes application programming interfaces; here one practice management system is named specifically and the rest are described generically as deep integrations. No published interfaces were located, so a customer wanting to build their own experience on this data has no documented route.

The eligibility network is the differentiator worth crediting and it is the piece competitors in patient access most often lack.

Graded B.

Nothing published. No hosting platform, no region, no residency option, no subprocessor list, no retention position and no availability commitment was located.

The distinctive element on this record is that a substantial part of the deployment is physical. Kiosks and tablets sit in waiting rooms, which means the vendor's software runs on hardware in the customer's premises, in a public space, unattended, holding a session containing clinical and payment data. Nothing published describes what is stored locally on those devices, how they are provisioned or updated, what happens to a device that is lost, stolen or decommissioned, or how a session is protected when a patient walks away mid registration. That is a deployment surface no other record in this session has, and it is entirely undescribed.

The voice agent adds a second undocumented path. Handling more than 50,000 calls a month requires telephony infrastructure and speech processing, and nothing states where either sits or who provides them.

Availability deserves specific mention because of what fails. If check in, eligibility and payments are down, patients queue in the waiting room and cannot be registered, and a practice that has reduced registration staff on the strength of this platform has no manual fallback at scale. The company markets operating with leaner teams, which makes uptime a clinical operations risk rather than an inconvenience, and no uptime commitment, status history or continuity position is published.

Graded D on the absence rather than on any evidence of a problem.

Nothing is published. No price, no unit of charge, no tier structure, no implementation fee, no contract term and no minimum was located. The site offers a demo booking incentivised with a gift card rather than any pricing information.

The packaging question is now larger than it was. With eight named agents plus the underlying registration, eligibility and payments products, and an explicit pitch that one platform replaces as many as ten point solutions, a buyer needs to know which agents are included in a base licence and which are separately priced. A consolidation argument is a cost argument, and it cannot be evaluated against a competitor stack without a number on either side.

The unit of charge is unstated and the candidates diverge sharply for this product. Per location, per provider, per kiosk or device, per patient visit, per call minute for the voice agent, or a percentage of collections would each produce very different economics, and the last of those matters especially because the payments agent collects money. Whether Clearwave takes a percentage of what it collects, a flat fee, or both is the single most consequential missing fact on this record, and it is the sort of arrangement that is common in payments and rarely volunteered.

One related disclosure does appear and is worth crediting as unusually candid: the payments agent is described as collecting credit card surcharge fees, and the company publishes commentary on card processing fees for specialty practices. Naming surcharges openly is more than most payment adjacent vendors do.

Graded D.

Deliberately concentrated on a segment, deep within it, and thinner outside it.

The core market is specialty practices and multi site groups, and the product is built around what makes those hard: complex provider scheduling rules, service line and provider specific intake workflows, and high volume front desks. Named examples run to orthopaedics, and the company describes scheduling logic developed specifically to handle the most complex provider requirements. Intake workflows presented by appointment service, provider and location is real specialty depth rather than a generic form builder.

Coverage extends upward and downward from there. Federally qualified health centres are addressed with their own material, hospitals and health systems are a stated segment with standardisation across locations and service lines as the pitch, and the eligibility product reaches more than 900 payers, which is national coverage on the insurance side.

Channel coverage is complete for the front desk: kiosk, tablet, mobile, pre visit link, inbound phone, online booking and search. The kiosk heritage means the physical waiting room is covered as well as the digital one, which no other record in this session addresses.

What is thinner is the enterprise end. Named integrations skew to practice management systems, with NextGen named specifically and marketed jointly, and no academic medical centre or large integrated delivery network deployment was located. Against Kyruus Health on the adjacent record, which reports 600 hospitals and names four record systems, this is a different and smaller market by design.

Graded B.