Laguna Health

Language models are the whole product. The platform listens to care management conversations, extracts the life circumstances and obstacles that do not appear in a record, generates the documentation, and scores whether the care manager pursued the clues a member offered.

The company is specific about the technology rather than gesturing at it, stating that it uses large language models and that it attends to the bidirectional structure of a conversation rather than treating it as a transcript to summarize. There is no content library, no rules engine and no meaningful software product underneath: strip out the models and what remains is a dashboard with nothing to display.

A no fabrication promise with nothing that lets anyone check it, and no review step before the output is written. The company's own blog carries a Q&A in which its vice president of conversational AI says the solution is "secure, backed by evidence, accurate, and hallucination-free". Hallucination free is a promise that nothing is added that the call did not contain, and a promise like that can hold only where the product makes additions checkable against their source. Nothing published here does: no link from a note or member update back to the conversation that produced it, and no review before content is written. A buyer who believes the promise will not staff the check that would catch an invented detail. The home page cuts the other way, reporting clinical content accuracy above 99 percent, itself an admission that some content is wrong, published with no method or denominator.

The workflow leans on the claim: member information is updated automatically after every interaction, post call work is handed to the AI, and no review before anything is written to the member record appears on the home or platform pages. The supervisory product, Laguna Insight, scores 100 percent of interactions to show team leaders where care managers need help, with nothing published on whether a care manager can see or contest a score.

The architecture is described in general terms and nothing is identified. The Laguna Reef platform page calls itself the underlying technology behind the products and describes four configurable layers: communication methodologies such as motivational interviewing and alliance building; compliance monitoring against member consent, HEDIS, NCQA and URAC standards; summaries in custom formats or standard ones such as SOAP and SBAR; and a dynamic member profile that detects and documents member barriers mapped to a third party ontology or to Z codes. The company's blog describes extracting about 50 psychological and social traits per member into a structured database.

No model, model family, training data, validation method or version practice is named on the home, platform or privacy pages, and "AI that's trained, secure, and compliant for healthcare" does not say trained on what. The accuracy figure on the home page, above 99 percent, comes with no description of how it was measured.

The architecture is described and no model provider is named. The privacy policy lists the categories of service providers that receive personal information (IT service providers, cloud storage, customer support, security monitoring, marketing and analytics) and says business associate agreements are in place wherever they access protected health information; the Laguna Reef platform page says data is stored and processed in United States based data centers. No speech recognition or language model provider, cloud host or subprocessor list is named on the home, platform or privacy pages.

The chain also runs outward. The policy permits aggregate, anonymized or de identified information about users to be shared with third parties for marketing, advertising, business, research or similar purposes, and permits de identified data derived from protected health information to be used for any lawful purpose. For a company that extracts psychological and social traits from recorded care conversations, what counts as de identified is a harder question than it is for a claims table, since voice and conversational detail identify people in ways structured fields do not.

Published peer reviewed research from the company, answering a different question from the product's headline claims. The publications page lists a randomized clinical trial Laguna co-authored in the Journal of Healthcare Management (2023, volume 68, issue 4): adults discharged home after a cardiovascular admission were randomized 1:1 to behavioral and emotional support from a coach (motivational interviewing, patient activation, attention to mental health and substance use) or to usual care, and observed readmission costs were significantly lower in the intervention group. A second peer reviewed paper, in Population Health Management, sizes the cost of post discharge readmissions, and an actuarial study with SBA Actuaries is listed.

That trial tested a human delivered care model, the one the software is built to support. The product's own headline figures on the home page, 66 percent less documentation time, clinical content accuracy above 99 percent, 96 more care plan enrollments per care team member a year, and a 26 percent efficiency gain in one month at myLaurel, come with no method, period or denominator, and the page credits its efficiency results to randomized trials and actuarial studies that measured readmission cost rather than efficiency.

Real disclosures on collection and de identification, and nothing on retention windows or training use. The privacy policy names call recordings among what is collected when a person communicates with a care team by chat, email, video or phone; says a health insurance carrier may append contact details, medical information and protected health information; uses the information for call analysis and summaries; permits de identified and aggregate data to be shared with third parties for marketing, advertising and research; and offers deletion on request, with the caveat that aggregate or anonymous derivatives may remain on company servers. The company's blog describes extracting about 50 psychological and social traits per member into a structured database.

No retention period is published for raw audio as distinct from derived text, nothing states whether member conversations train models, and nothing says what a member is told about recording and software analysis, in a country where several states require every party's consent to record a call. No safety engineering is published either: how a mis-extracted trait or a wrong summary is caught and corrected.

Laguna states its business associate status in a substantive privacy document, and the agreement and its scope are not fully published. The privacy policy says that where a member's health plan or provider is a covered entity and Laguna provides services on its behalf, Laguna may qualify as its business associate; that protected health information collected in that role is used and disclosed only as HIPAA permits; that business associate agreements are in place with service providers who access it; and that a deletion request covering protected health information can be sent to connect@lagunahealth.com. No business associate agreement is published.

Two sentences in the same policy narrow the protection when read together. Information collected when a person signs up directly, outside a provider or insurer relationship, is not protected health information. And a member who signs a separate HIPAA authorization releasing health information to Laguna moves that information out of HIPAA and under the terms of the authorization. The same member's health facts can therefore carry different protection depending on the route by which they arrived, and the member exposed is one who engages directly or signs the authorization without knowing what it changes.

A recognized attestation is displayed without its detail. The site footer carries a compliance row with a HIPAA logo and an AICPA logo, the mark of the body that governs SOC examinations, presented as the company's own. It does not say which report it refers to (SOC 1, 2 or 3), whether it is Type I or Type II, what period it covers or who performed it. The Laguna Reef platform page describes the controls behind it: data stored and processed in secure United States based data centers, 256 bit encryption in transit and at rest, single sign on, and governance controls customizable to each organization. No trust portal or downloadable report is linked.

That leaves the central question open: whether this was a point in time design examination or one testing whether controls operated across a period.

No clearance, authorization or submission is claimed, and none is needed. Supporting a care manager's conversation, drafting documentation and surfacing obstacles is not device functionality. The regulator that matters is not the FDA: the company names Medicare star ratings and HEDIS quality measures as explicit value drivers, which places the product inside plan quality and payment rules administered by the Centers for Medicare and Medicaid Services. A device framework is the wrong body of law for this product, and the live questions concern documentation integrity and quality measure attribution rather than clinical safety review.

Responsible AI is committed to in policy language with no evaluation behind it. The Laguna Reef platform page says the company's enterprise grade AI is responsible, secure and trustworthy and that it applies that standard to everything from technology to business practices, and lists governance controls that meet leading industry requirements and can be customized to each organization. No bias evaluation, subgroup performance, model card or named governance process appears on the home, platform or privacy pages.

Two exposures are specific to this product. Speech and language model performance varies with accent, dialect, first language, age and emotional state, and the members on these calls are disproportionately post discharge, complex care and socially burdened. And Laguna Insight scores care managers on their interactions, so a model that hears some members less reliably registers fewer barriers in those calls and can mis-score the care managers assigned to them: model bias becomes staff evaluation. The company also extracts psychological traits about members into a structured database, which is inference about people that deserves its own review.

Nothing published on what happens when the system is wrong. The home page, the Laguna Reef platform page and the privacy policy publish no warranty, indemnity, remediation commitment or route for a member to challenge what was recorded about them, and none describes a mechanism for challenging an output, such as source links from a note back to the call or a review step before the record is updated.

The one performance figure attached to the output is clinical content accuracy above 99 percent on the home page. It is published with no method, denominator, sample or definition of clinical content, so it is not an error rate the company can be held to.

The exposure is concrete: generated documentation and automatically updated member information feed care plans, quality reporting and, through Laguna Insight, the scoring of care managers themselves, so an error propagates in two directions.

Integration is described as a choice and no system is named. The Laguna Reef platform page offers light integration for a fast start or deeper connection to an existing CRM, and the company's blog describes its tools as working across cloud based systems; the home page says member information is updated automatically after every interaction, and the site offers a Chrome extension. No record system, care management platform, CRM, telephony platform or interoperability standard is named on the home, platform or privacy pages, though the member profile maps barriers to Z codes or a third party ontology, which is a coding standard rather than a connection.

The care management systems this product must sit beside are a fragmented market of their own, and a browser extension that reads and writes across them depends on which ones it supports.

The hosted option and its residency are stated, and the processing path is left open. The Laguna Reef platform page says data is stored and processed in secure United States based data centers, with encryption in transit and at rest and single sign on, and describes integration as a choice: light integration for a fast start, or deeper connection to an existing CRM. The privacy policy separately says personal information is transferred outside the country where it was collected only to countries with adequate protection, which leaves open whether any processing, including any external language model service, happens outside the United States.

The material question for a conversation product is audio: a voice recording identifies the speaker independently of its content, and nothing on the home, platform or privacy pages says whether raw audio is retained after transcription or only derived text, or whether customers are isolated from one another.

Every route to a price runs through a sales conversation. The home and platform pages publish no price, unit of charge or contracting model; the routes in are a demo request, a free trial request form, and a downloadable one pager and impact summary behind a form. The basis question is material for a product sold to health plans and to outsourced care organizations: per care manager seat, per member under management, per minute of analyzed conversation or per program would produce very different bills from the same deployment.

The value case the site does publish (96 more care plan enrollments per care team member a year, 66 percent less documentation time) is framed per care manager, which suggests but does not state a seat basis.

Focused on care management conversations rather than a disease, which is a coherent scope. The company's origin is the transition out of hospital, where its published trial worked with cardiovascular discharges, and the product now covers care management and coaching and training across two buyer types named on the site: health plans, and virtual care organizations including outsourced service providers, third party administrators and care navigators. The setting is the member's home and the telephone rather than a clinical facility, and the Laguna Reef platform configures the product to each organization's methodologies, from motivational interviewing to quality standards such as HEDIS, NCQA and URAC.

What it lacks is depth. Coverage is United States only, and the one customer named with a result on the home page is myLaurel.