Attune

The model is the entire product and there is nothing else. An autonomous conversational agent places and answers calls, understands what a patient says, plans a sequence of actions toward a goal, operates other systems to carry them out, and decides when to hand off. Remove it and no product remains.

The company is also accurate in how it distinguishes itself, describing the difference between conversational systems that follow rigid scripts and an agent given a goal and latitude over how to reach it, which is a real architectural distinction rather than a marketing one.

The highest autonomy product assessed in this segment, and the oversight architecture around it is not published. The agent initiates outbound calls on its own, holds unsupervised conversations with patients and members, books against live schedules and completes health assessments, described by the company as happening without direct human intervention. Escalation exists and carries full conversational context to a person, which is the right design.

What is absent is everything a buyer needs to trust it: the criteria that trigger escalation, what the agent does if a patient discloses a clinical emergency or acute distress mid call, whether and how the agent identifies itself as artificial rather than human, and what supervision exists over a conversation no person hears.

A comparable company in this same segment publishes a specific answer, describing several specialised agents running on every call with some dedicated to safety supervision and real time observability for the operator. That is the standard this record is graded against, and it is a published standard rather than a hypothetical one.

Two retrieval passes located no architecture description, no statement of which models are used or whether they are built or licensed, no validation methodology, no accuracy figures and no publication. The word agentic carries the whole technical claim.

For a voice product the specific unpublished number is speech recognition and understanding accuracy on the actual call population, because everything downstream depends on it: an agent that mishears a medication name or a symptom does not fail visibly, it proceeds confidently on the wrong understanding. Nothing addresses it.

One case is offered and it is unnamed and unquantified: a major health plan is described as deploying the agent across its entire Medicare member population to complete health risk assessments, with the outcome given as more assessments completed than previously possible, which supplies no baseline, no denominator and no completion rate. Two passes located no named customer, no peer reviewed publication and no independent evaluation.

The company launched publicly in January 2026, so at the time of this assessment it is roughly six months old, and a thin evidence record is expected rather than surprising. Recorded so the grade is read as a description of what exists today rather than as a judgement about the product.

Two retrieval passes located no retention position, no minimisation statement, no consent position and no policy on secondary use. The gap matters more here than for a passive analytics tool, because this system generates the conversation as well as processing it: it is placing calls to patients who did not ask to be called, recording what they say, and writing results back into systems.

Recording consent, disclosure that the caller is artificial, and what happens to the audio afterward are the three questions a patient would ask if anyone asked them, and none is answered in anything published.

Two retrieval passes located no HIPAA statement, no Business Associate Agreement terms and no privacy or legal page. Graded on published posture. Agreements certainly exist for the health plan deployment described. Worth noting alongside this that outbound automated calling carries a second body of law beyond health privacy, since telephone consumer protection rules govern automated and prerecorded calls and have been the subject of enforcement against healthcare callers, and nothing published addresses how the product handles consent for automated outbound contact.

Two passes located no SOC 2, no HITRUST, no ISO 27001, no trust centre and no vulnerability disclosure policy. For a company at this stage that is unsurprising, and it is also the axis most easily and quickly corrected: a competitor in an adjacent segment founded in 2024 has completed and published an audited report, which shows that youth is not itself the obstacle. A health plan handing an outbound calling system its entire Medicare membership will require an attestation before go live, so the document likely exists privately.

No clearance, authorisation or submission located and none needed. Scheduling appointments, closing care gaps and administering a structured questionnaire by telephone is administrative work and sits outside device regulation. As with the rest of this segment, the regulator that matters is elsewhere: the health risk assessment use case places this product inside Medicare Advantage payment and quality rules administered by the Centers for Medicare and Medicaid Services, and automated outbound calling sits under telephone consumer protection rules. A buyer assessing this against a device framework would be examining the wrong law entirely.

Nothing published, and two exposures are specific enough to name. The first is speech performance in the population being called. The flagship deployment works through an entire Medicare membership, which is an older population where hearing impairment, slower turn taking, dysarthria and regional or non native accents are all more common, and these are precisely the conditions under which speech recognition degrades.

The company states that the agent supports diverse patient populations with human like tone and empathy, and publishes no performance data behind it. This is now the fourth record in this index to carry the same unexamined exposure. The second concerns the health risk assessment use case.

Assessments conducted outside a treating provider encounter have attracted sustained federal scrutiny because of their role in generating risk adjusted payment from diagnoses not otherwise documented, and an autonomous agent administering them at population scale intensifies rather than resolves that question. Nothing published describes what the completed assessment is used for, whether its contents can support a diagnosis code, or what controls separate care gap closure from payment. That should be established in writing.

The product must integrate to function at all, since booking an appointment requires reading a live schedule and writing to it, and the company describes reading and writing through interfaces, webhooks and secure transfers. One capability is unusual and worth noting: the agent navigates phone trees and manages being placed on hold, which means it can operate as a caller into organisations that offer no interface at all, substituting a phone call for an integration where none exists. That is a pragmatic answer to a real problem in this market. Held at C because no record system, scheduling platform or telephony partner is named anywhere located, so the depth is entirely asserted.

A hosted voice platform, necessarily, since the agent must sit in the telephony path to place and receive calls. Two passes located no architecture description, hosting region or residency commitment. The question that matters for this product class is whether call audio is retained or only derived transcripts and structured results, because a voice recording identifies the speaker independently of what was said, and a system calling an entire Medicare membership accumulates a very large volume of it. Nothing published distinguishes them, and nothing describes whether any part of the speech or language pipeline runs through an external model provider.

No pricing, pricing mechanism, basis of charge or funding position located in two passes, which is a fuller absence than usual since even the round that funded the launch was not identified. The basis of charge is a live question for autonomous voice, where per call, per minute, per completed task and per member produce very different bills and shift the risk of a long or failed call between the parties.

For a company six months into public operation being asked to carry outbound contact with an entire Medicare membership, a buyer should establish funding position and runway directly, on the supplier continuity ground this index applies from the Behold.ai precedent.

Function led rather than specialty led, covering appointment scheduling and confirmation, follow up after care, care gap closure, and structured health assessments, across both provider and health plan buyers. It is not tied to any clinical specialty, which is appropriate since the work is engagement rather than care. Setting is the telephone and therefore the patient's home.

Multiple languages are implied by the claim to serve diverse populations though no count is given, which is a material difference from a competitor in this segment that states its language coverage explicitly. Geography is the United States. Held at C because breadth of function is not the same as depth in any of them, and at six months old none of the named use cases is evidenced beyond a single unnamed deployment.