Humata Health

Two assets carry this product and only one of them is a model. Deciding whether a given order requires authorisation, then finding and assembling the specific clinical evidence a specific payer needs to approve it, is genuinely a reasoning task over an unstructured chart.

The other asset is the network. The company's own framing is that the hardest part is making systems at every payer talk to systems at every hospital, and it describes building that connectivity and holding the largest ecosystem of integrations in the category. That is the moat is the network case this index applies elsewhere, and it is why the grade is B: a competitor with an equal model and no connections could not deliver the same outcome.

The demonstrated position is high. At a named health system the company reports authorisation running from order to approval with no human intervention at any point, which is full automation of a process that normally consumes a dedicated department.

What holds it at B is that the claim is anchored to a first rather than to a rate. Nothing published states what share of requests complete touchless across the customer base, at what confidence the system stops and hands to a person, or what happens when the assembled documentation is insufficient. A category whose failure mode is a patient waiting needs those numbers, and the comparison record in this index for high autonomy publishes an automation rate and contracts to it.

Worth noting on the correct side of the ledger: the automation here assembles a case for approval rather than issuing a denial. The consequential decision still sits with the payer.

The workflow is described clearly and the technology is not. No model, architecture or method is named, no accuracy or completeness figure is published for the documentation the system assembles, and the claim of the largest integration ecosystem in the category carries no count of integrations or payers.

That matters more than usual because the product's core act is a judgement: deciding which parts of a chart constitute evidence that a specific payer's criteria are met. A buyer cannot tell from public material how often that judgement is right, only that at one health system it was right often enough to run untouched.

No retention schedule, encryption detail, model training position, hosting arrangement or sub processor list was located. The data movement is worth naming because it is the point of the product rather than a side effect: the system reads the chart to find supporting evidence and then transmits that clinical documentation to a payer, so its function is to move patient information across an organisational boundary, from an entity the patient chose to one they did not.

How much it sends, whether it sends more than the criteria require, and what the payer retains are all questions a provider should ask, and the incentive runs the wrong way. Over disclosure in pursuit of approval is a real failure mode rather than a hypothetical one, because more supporting evidence makes approval more likely and nothing in the workflow rewards restraint, so a system optimised for approval rate will tend to send more of the chart than the criteria strictly need.

The minimum necessary principle is precisely what that pressure erodes. Nothing published describes whether the assembled packet is scoped to the criteria, whether a human reviews what is being sent before it goes, or what the payer is permitted to do with clinical narrative it received for an authorisation decision. Ask all three, plus retention on the assembled packet at both ends.

Corroboration is unusually strong and measurement is absent, and this index does not let the first substitute for the second.

What exists: a named health system reported as the first in the country to run fully touchless prior authorisation, selection by the federal Medicare agency as a technology partner for a live national model, participation in a federal acceleration initiative, and an integration with a major utilisation management business. Selection by a public agency after its own process is a meaningful external signal, and stronger than most vendor references here.

What is absent: any published rate. No touchless percentage, no turnaround time distribution, no approval rate comparison against the manual process, and no study. The company's own goal, that the industry return more than 80 percent of decisions in real time by 2027, is a target rather than a result.

Graded on an honest basis. No retention schedule, encryption detail or model training position was located in this pass.

The data movement is worth naming because it is the point of the product. The system reads the chart to find supporting evidence and then transmits that clinical documentation to a payer, so its function is to move patient information across an organisational boundary. How much it sends, whether it sends more than the criteria require, and what the payer retains are all questions a provider should ask, since over disclosure in pursuit of approval is a real failure mode rather than a hypothetical one.

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

The contracting picture is more layered than usual because the company serves providers, integrates with a utilisation management business, and acts as a technology partner to a federal agency. Those are three different relationships with three different sets of obligations, and which governs a given data flow is not described publicly.

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.

Selection as a technology partner in a federal Medicare model implies security review by the agency, which is a demanding gate, and nothing about it was retrieved.

No device pathway applies and none is claimed. Assembling and submitting an authorisation request is administrative.

The regulatory environment is nonetheless the most consequential of any record in this category, and it changed in January 2026. Federal interoperability and prior authorisation requirements now mandate electronic submission and faster turnaround, which is what makes this segment move. Separately, the Medicare agency's own model applying artificial intelligence to clinical review went live in the same month with this company as a technology partner.

So the buyer is purchasing into a process that is being restructured by regulation while they buy, and the vendor is helping build the infrastructure the regulation assumes. That is a strong commercial position and a reason to read the vendor's public statements about the direction of reform as those of an interested party.

Nothing published on evaluation, monitoring, error rates or subgroup performance, and the position this company occupies makes that gap the most important thing on the record.

The company's provider facing identity is built for yes, and the automation it sells assembles a case for approval. It is simultaneously a technology partner in a federal Medicare model named for the reduction of wasteful and inappropriate services, which applies artificial intelligence to clinical review. Those are not contradictory, and they are not the same job either: one accelerates approval, the other exists to decline what should not be paid for. The same capability serves both.

Nothing published describes how the company separates them, what governs the model on the federal side, or how a patient or clinician would learn that an automated review contributed to a decision about their care. Given that this index has already recorded several states restricting artificial intelligence driven coverage denials, a vendor operating on both sides of that line should be expected to say more.

The workflow is described clearly and the technology is not: no model, architecture or method named, no accuracy or completeness figure for the documentation the system assembles, no evaluation methodology, and no warranty, indemnity or remediation commitment, while a claim to the largest integration ecosystem in the category carries no count of integrations or payers. That matters more than usual because the product's core act is a judgement rather than a transfer.

It decides which parts of a chart constitute evidence that a specific payer's criteria are met, which is an interpretive task with two distinct failure modes: assembling a packet that omits the element the payer needed, producing a denial that looks like a clinical disagreement rather than an assembly error, or asserting that criteria are met on evidence that does not support it.

A buyer cannot tell from public material how often that judgement is right, only that at one health system it was right often enough to run untouched, and a single site's tolerance is not a performance figure. The affected party is a patient whose authorisation was delayed or denied, and who will never learn that the packet rather than their clinical picture was the problem.

Ask for the approval rate on submissions the system assembled unaided compared with human assembled ones, the rate of denials attributable to missing documentation, and what the system does when it cannot find supporting evidence.

Interoperability is the product rather than a supporting feature. The workflow starts when a clinician enters an order in the record system and ends with a payer decision returning, which requires live connection at both ends, and the company describes the payer to hospital connectivity problem as the central obstacle it is solving.

Distribution reinforces it: the technology is available within a widely deployed ambient clinical assistant, so authorisation can be initiated from inside the documentation workflow, and an integration with a utilisation management platform extends payer reach. Held at B because no named record vendor certification, interface standard or count of payer connections was located.

Not described. No hosting model, region or retention position was located.

A product that transmits clinical documentation to payers and to a federal programme is moving regulated data between several organisations, and where it rests in between is not published.

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

The unit question is sharper here than usual because the company now sells to two very different buyers. A large health system replacing an authorisation department has a different economics from an independent practice where one person handles prior authorisation between other duties, and the new provider portal is aimed squarely at the latter. Per request, per provider and per site would each land very differently across that range. Ask which applies, and ask what happens to the fee when a request is denied, since the vendor is paid for work whose value to the provider depends on the answer.

Broad within one process and deliberately widening. The company states coverage of all services and procedures rather than a subset, which distinguishes it from vendors handling only specialty drugs or only imaging, and prior authorisation is specialty agnostic by nature.

The buyer range expanded materially in June 2026, from large health systems and provider groups to independent practices, regional health centres and specialty clinics, which are the settings where this work is still done by fax and telephone. Reach extends to payers through the utilisation management integration and to Medicare through the federal model. Held at B because everything sits inside one administrative process and coverage is United States only, which is inherent since prior authorisation is a feature of this payment system.