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.

Held below A because the output 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.

Two quite different autonomy positions sit in one company and a buyer should separate them.

The data work is automated at scale: records are validated, corrected and enriched on a recurring cadence without a person reviewing each one, which is the only way the volume is tractable. Nothing published describes a confidence threshold at which a record is routed to a human.

The outreach product announced in August 2026 goes further. An agent telephones a provider's office to confirm details, which means an automated system now initiates the contact that resolves the ambiguity rather than only inferring across sources. The company frames this as beginning from trusted data so calls are targeted confirmations rather than open discovery, which is a sensible constraint. What is not published is what the agent does when the person who answers contradicts the record.

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 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. That is rare in administrative software.

What is absent is accuracy expressed as a rate. No precision or recall figure for the corrections themselves was located, and a claim of processing clinical data 100 times faster than people is a speed claim carrying no accuracy denominator.

One distinction materially lowers the risk on the flagship product and should be stated rather than assumed, because a reader scanning this index would otherwise apply the wrong frame. Provider directory data is business information about professionals, covering their names, locations, specialties, contact details and network participation, and it is not patient information, so the core product carries a far smaller privacy exposure than most records here.

A vendor whose primary holding is professional business data should not be graded as though it held charts. What changes the picture is where the platform extends. The company states it handles clinical results and value based care data and is expanding into medical record data, and those are patient information in the ordinary sense, governed differently and carrying the full set of questions this axis asks.

No retention, encryption or model training position was located for either side, and nothing separates the two: a buyer cannot tell whether the controls described for directory work extend to clinical content, or whether the same models and infrastructure serve both. Nothing else is enumerated either, with no hosting arrangement or sub processor list found.

Ask which products touch patient data, whether the infrastructure is shared with the directory product, retention for clinical content specifically, and for a sub processor list covering the clinical side.

The strongest evidence position of any administrative vendor in this index, and it is earned in an unusual way.

The company's data was not merely validated by outside researchers; it was the instrument in peer reviewed research published in a leading medical journal, which found that more than four in five physicians listed in insurer directories had inconsistent entries. That work was then referenced in a MedPAC report to Congress. So the platform was accurate enough to be trusted as the measuring device in published research on a national policy question, and the finding entered the policy record.

That is the same shape this index credited when a vendor's software was the selection instrument in trials that changed treatment guidelines: contributing the instrument outranks publishing your own accuracy. Deployment corroborates it, across most states, national insurers and a top ten health plan.

One honest qualification. The published work establishes the scale of the problem the company sells into. It does not by itself establish how much of that problem the product fixes, and no before and after directory accuracy measurement was located.

Graded on an honest basis, with a distinction that materially lowers the risk on the core product and should be stated rather than assumed.

Provider directory data is business information about professionals, their names, locations, specialties, phone numbers and network participation. It is not patient information, so the flagship product carries a far smaller privacy exposure than most records in this index.

That changes where the platform extends. The company states it handles clinical results and value based care data and is expanding into medical record data, and those are patient information in the ordinary sense. No retention, encryption or model training position was located for either side.

Graded on an honest basis. No compliance statement or agreement posture was located in this pass.

Agreements clearly exist with national insurers and a top ten health plan. The point worth establishing is which product a given agreement covers, since the provider data work and the clinical data work sit under different obligations, and a plan contracting for directory accuracy is in a different position from one sending clinical results.

Recorded honestly and provisionally: the dedicated trust and security search this index requires was not run in this pass, and no attestation was encountered incidentally.

A vendor operating inside national insurers and a top ten health plan has passed substantial security assessment, so the absence reflects what was retrieved rather than what exists.

No device pathway applies and none is claimed. Cleaning administrative data is not a clinical determination.

The regulatory context is nonetheless the reason this product exists, and it is unusually specific. Federal law requires health plans to verify provider directory information at least every 90 days and to act on updates within two business days, and a subsequent Medicare Advantage statute narrows the focus to directory accuracy with civil penalties for repeated inaccuracies. State regulators can fine plans directly. So the buyer is purchasing compliance with a named and enforceable obligation rather than an efficiency, which is a different commercial position from most of this index and one a buyer can size precisely.

No subgroup analysis, error rate or monitoring policy was located, and the distributional question here is sharper than the abstract framing suggests.

Directory errors are not evenly distributed. They concentrate among small independent practices, behavioural health clinicians and providers outside large integrated systems, because those are the practices without the administrative staff to keep rosters current. Those are also precisely the clinicians patients struggle most to reach, which is why the phenomenon has a name and has drawn congressional attention in mental health specifically.

A resolution system works best where source signals are abundant and consistent, which describes large systems rather than solo practices. If accuracy improves fastest where it was already best, the directory becomes more reliable in exactly the parts of the network patients had least trouble with. Nothing published reports accuracy by practice size or specialty. That single breakdown would be the most informative disclosure the company could make.

The method is described in peer reviewed literature, which means an outside reader can examine how the entity resolution works rather than taking a platform claim on trust, and that is rare in administrative software where the usual standard is a marketing page.

Around it the operational scope is stated with unusual precision: the number of attributes validated per provider, cross checking across thousands of sources, the refresh cadence, the ability to process ad hoc batches, and coverage quantified as a share of the country's clinicians. A buyer can tell what is checked, how often and against how much. Held at C because accuracy is never expressed as a rate.

No precision or recall figure for the corrections themselves was located, and a claim of processing clinical data a hundred times faster than people is a speed claim carrying no accuracy denominator. Precision is the number that matters for this output, because the product's job is to change a record: a wrong correction to a directory entry sends a patient to a clinician who does not practise there, or marks a practising clinician as out of network, and the person who suffers is a patient who never sees the system that caused it.

Recall matters equally, since an uncorrected stale entry persists silently. No warranty, indemnity or remediation commitment was located. Ask for precision and recall on corrections, and what review sits between a proposed correction and a published directory.

The relevant systems are payer side, provider data management, credentialing and directory platforms, rather than the electronic health record, and the record is graded on that basis rather than penalised.

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. Stated expansion into medical record data would change this axis materially and has not yet. No interface standard or named platform integration was located.

Described only as a cloud platform. No hosting region, retention schedule or customer controlled option was located.

The company operates across four continents with a substantial engineering presence outside the United States, which for payer data is a question a buyer will ask directly and which public material does not answer.

Nothing published: no price, no mechanism, no unit of sale.

The unit is the interesting question here because the plausible bases differ sharply. Per provider record under management, per attribute validated, per outreach call completed and a flat enterprise licence would each behave very differently for a plan whose network is growing. The outreach product adds a second dimension, since a call that reaches nobody still costs something to attempt.

Against that, the value is unusually easy to size, because the obligation is statutory and the penalties for repeated directory inaccuracy are defined. A buyer can price this against a known exposure rather than against a productivity estimate.

Wide on the payer side and specialty agnostic by construction, since a directory contains every clinician a plan contracts with.

Stated reach is substantial: live in most states, analysing data covering more than 80 percent of United States clinicians, with national insurers, a top ten health plan and a national behavioural health organisation among named deployments. Behavioural health is a pointed inclusion given that directory inaccuracy is worst in that specialty.

The product set extends beyond directories into clinical results, payment accuracy and value based care, so the company is broadening from one data domain into several. Held at B because the buyer is almost exclusively the health plan, with providers appearing as the subject of the data rather than as customers, and because coverage is United States only.