HiLabs

The models do the work. Reconciling conflicting statements about a clinician across thousands of sources into one record that is probably correct is an inference problem, not a matching problem, and the company positions explicitly against automation that simply propagates existing errors faster.

The output also compounds into an asset that is not a model. A validated view of more than 80 percent of United States clinicians becomes the reference the next inference is checked against, and a competitor with equally good models and no accumulated corpus would resolve less confidently. That is the moat is the dataset case, and here it is a direct consequence of the models rather than a substitute for them.

Automation is described across several products, and one of them promises nothing is missed. Directory records are validated, corrected and enriched on a recurring cadence without a person reviewing each one, with no published threshold for routing a record to a human; the outreach product's agent telephones provider offices to confirm details; and the value based care suite recomputes settlements before payment and ranks anomalies for review, so a person sees the flagged cases before money moves.

The value based care page describes CareIntel as an action driven care optimization platform "ensuring no signals are missed and enabling proactive intervention". That is a claim about what is absent, which no review of the surfaced signals can show, so it grades here whatever the other products' designs: a care team told nothing is missed will not look for the member the panel view never raised. The page marks CareIntel as evolving, which does not qualify the sentence.

Operationally specific in the ways that matter to a buyer. The company states the number of attributes validated per provider, more than 60, describes cross checking across thousands of sources, states the refresh cadence and the ability to process ad hoc batches, and quantifies coverage at more than 80 percent of United States clinicians; the home page describes healthcare trained AI that replicates subject matter expert behavior, geospatial intelligence and plug and play interoperability.

The stronger disclosure is methodological: the approach has been described in peer reviewed literature, which means an outside reader can examine how the resolution works rather than taking a platform claim on trust.

What is absent is accuracy expressed as a rate. The company publishes no precision or recall figure for the corrections themselves, and a claim of processing clinical data 100 times faster than people is a speed claim carrying no accuracy denominator. The newer outreach and value based care products name no model or version.

The host is named, and no model or voice provider is. The company's announcement of its third HITRUST i1 certification, as reported, places the MCheck platform on AWS cloud infrastructure, with offices in Bethesda, Pune and Bangalore inside the certified scope. The platform now includes an agentic outreach product that calls provider offices and a value based care suite with plain language chat over care data, which implies language and voice models; which providers supply them is not stated on the home page, the value based care page or the September 2026 certification announcement, and no subprocessor list is published.

Recording the host matters for residency and breach scope; it does not say which model sees a member's or provider's data. Directory work is business information about clinicians, while the clinical, value based care and outreach products handle member and patient data, so which parties touch which product is the question.

Published research on the problem, vendor figures on the product, and no measurement of how much the product fixes. The company says its method was the measuring instrument in peer reviewed research published in JAMA, which found more than four in five physicians listed in insurer directories had inconsistent entries, and that the work was cited in a MedPAC report to Congress; the home page restates that 81 percent of provider directory data is incorrect. Its own figures are scale and output: more than 40 billion health data records analyzed, 100 million directory records reviewed and 25 million errors corrected, and for value based care more than $1.5 billion of annual payments monitored and $80 million flagged for review. It says it serves six of the ten largest United States health plans, none named.

The headline claim is that the AI corrects directory data the way a subject matter expert would, and the published research answers a different question, how bad directories are, not how accurate the corrections are. No precision, recall or before and after directory accuracy is published, and the errors corrected figure has no denominator or check behind it.

Security assurances, and nothing on what happens to member data. Provider directory data is business information about clinicians, their locations, specialties, contact details and network participation, not patient information, so the flagship product carries a smaller privacy exposure. The rest of the platform does not: MCheck Clinical extracts clinical data, the value based care suite tracks member attribution and predicts which members will leave, and the outreach product calls provider offices on a plan's behalf. The HITRUST e1 and AI Security certifications address how data and models are protected from attack.

No retention period, training use statement or de identification step is published on the home page, the value based care page or the certification announcement, and the site has no privacy policy. Whether member and clinical data train the company's models, and whether directory and clinical products share infrastructure, is not stated. No safety engineering is published beyond the AI security controls.

No statement of status and no privacy document is published. The home page, the value based care page and the September 2026 certification announcement carry no HIPAA status, business associate agreement or privacy notice, the footer links no privacy policy or terms, and the site's privacy policy address returns a page not found. The HITRUST certifications are security attestations and do not state a HIPAA status (they are credited on Security). A 2021 post by a compliance software vendor announcing that HiLabs completed its HIPAA program is that vendor's marketing, not a statement on HiLabs' own estate.

The gap matters more as the product set widens. Directory work is business information about clinicians, but MCheck Clinical extracts clinical data, the value based care suite tracks member attribution and churn, and the outreach product calls offices on a plan's behalf; those handle protected health information, and a plan cannot read before contracting what governs it.

Recognized certifications are named as the company's own, with scope stated for one of them. The home page carries a SOC 2 logo and HITRUST e1 and HITRUST AI Security certification marks, and the company's September 2026 announcement says both HITRUST certifications cover its MCheck platform, the AI certification addressing risks such as prompt injection, model inversion and data poisoning. An earlier company announcement, as reported, describes a third consecutive HITRUST i1 certification covering the platform, its AWS infrastructure and its offices in Bethesda, Pune and Bangalore.

No report or letter can be opened from the site, the SOC 2 type and period are not stated, and the current announcement leads with e1, the essentials assessment, where earlier years were certified at i1; which assessment is current for the platform is not stated. No penetration testing statement or trust center is published.

No device claim is made, and the products are administrative and population health tools. Directory, roster, contract and network products clean and reconcile administrative data; the value based care suite audits attribution, reconciles settlements, predicts member churn and surfaces panel priorities for care teams through CareIntel, which the page marks as evolving. Ranking members by time sensitive panel priorities for a care team's intervention is population health work, and no treatment recommendation or diagnosis is marketed.

The regulation that bites is specific. Federal law requires plans to verify directory information at least every 90 days and act on updates within two business days, Medicare Advantage rules attach penalties to directory inaccuracy, and value based care settlements answer to CMS program rules, which the CMS Sentinel product tracks. The products answer to named obligations.

Policy language and AI security controls, and no evaluation of how the models behave. The September 2026 announcement frames the HITRUST AI Security Certification as strengthening AI governance, but what it validates is protection against AI specific attacks such as prompt injection, model inversion and data poisoning, which is security work. The value based care page describes outputs as traceable and reproducible. No governance framework, evaluation results or subgroup analysis is published on the home page, that page or the announcement.

The distributional question is sharp. Directory errors concentrate among small independent practices, behavioral health clinicians and providers outside large systems, the practices without staff to keep rosters current and the clinicians members struggle most to reach. A resolution system works best where source signals are abundant, so if accuracy improves fastest where it was already best, the directory becomes more reliable where members had least trouble. The same applies to the value based care suite's churn predictions and attribution audits across member groups.

Mechanisms exist to check an output, and nothing stands behind it. The value based care page describes an audit ready design in which every output is traceable, defensible and reproducible, settlements recomputed independently before payment, anomalies ranked by dollar impact and root caused automatically, and corrected payments sent with shared evidence to the provider; directory corrections are checked across thousands of sources on a recurring cadence. Those give a reviewer ways to challenge an output.

No warranty, indemnity, accuracy commitment or remediation obligation appears on the home page, the value based care page or the certification announcement, the site publishes no terms or privacy policy, and no precision figure for directory corrections or payment flags is published. The value based care page's statement that every product pays for itself in the first year is a model on a reference payer, not a commitment the customer can hold. A wrong directory correction sends a member to a clinician who does not practice there, and a wrong settlement lands on a provider group; neither has a published route back to the vendor.

The relevant systems are payer side, provider data management, credentialing and directory platforms, rather than the electronic health record.

The notable integration is with the industry provider data utility, whose collected data the company combines with its own models, which is the right architecture: use the existing attestation channel as one signal among thousands rather than treating it as truth. The company has stated plans to expand into medical record data, which would change its integration picture materially; that has not happened yet. The site names no interface standard or platform integration.

A single hosted platform on a named cloud, with location implied. The company's announcement of its third HITRUST i1 certification, as reported, describes the MCheck platform on AWS cloud infrastructure, with its offices in Bethesda and in Pune and Bangalore, India, inside the certified scope; the September 2026 announcement says the platform integrates with plans' credentialing, provider data, claims, network management and clinical data systems. No region, tenancy model or residency commitment is published on the home page, the value based care page or that announcement.

Where member and clinical data rest is implied by a United States company on AWS rather than committed, and with the India offices inside the certified scope, nothing says whether any data is accessed or processed from outside the United States.

A reference price is published for one product family, without tiers. The value based care page models the full four product VBC CareSphere suite for a reference payer (2.5 million members, 1.5 million value based care attributed lives, 450,000 Medicare Advantage lives) at a $3.9 million first year investment on existing data, against $13.6 million returned in year one and a 6.5 times net return over three years, with individual products said to deploy in three to seven months. A ContractsAI return calculator sits in the footer. A plan of a different size can scale the reference figure to size its own cost before contact.

What a buyer cannot establish: the unit of charge behind the reference figure (per member, per attributed life or per product), and any price for the directory, roster, network, outreach or clinical products. The return figures are the company's model, not a commitment.

Wide on the payer side, with validation behind part of it. HiLabs sells to Medicare, Medicaid and commercial health plans across provider data (roster automation, directory accuracy, network intelligence, contracts, provider outreach), clinical data extraction and value based care. The value based care suite covers the contract lifecycle: MCheck VBC CareSphere simulates value based contract terms before go live, audits member attribution, reconciles settlements against attribution and contract factors, predicts member churn and tracks CMS rule changes against a plan's book, with the company citing more than $1.5 billion of annual value based care payments monitored across 16 million lives. The company says it serves six of the ten largest United States plans across all lines of business in all 50 states, and a national behavioral health plan selected it for network adequacy.

No customer is named, the value based care figures are scale rather than results, and two of the four value based care products are marked as evolving. Coverage is United States health plans, with providers the subject of the data rather than customers.