Evidently

The engine is the product. Evidently reads structured and unstructured content across the record, including notes, labs, imaging, faxes, scanned documents and outside records from health exchanges, and the company describes an underlying encoding of millions of medical concepts and the connections between them. Every commercial surface, summaries, drafts, chat and reporting, is generated from that layer rather than bolted onto a system of record the company already sold. This is not the moat is the dataset case: the ontology exists to make the generative product work rather than standing in for it.

The widest generative surface in this category paired with the least described oversight, which is the pattern this index has graded C before. The product drafts admit notes, discharge summaries, emergency medicine notes and payer denial appeals, answers open clinical questions conversationally inside the EHR, and generates summaries from prompts the clinician writes. No confidence signal, no routing threshold, no abstention behaviour and no described review gate was located for any of it.

The custom prompt path deserves separate weight and is the reason this sits at C rather than B. When the user composes the request, no vendor side validation can cover the output, because the vendor never saw the prompt. Whatever testing exists applies to the summary types Evidently designed, not to a question a clinician invents at the bedside, and the organisation inherits an evaluation burden that is nowhere disclosed.

Ask which outputs have been validated, whether custom prompt output is treated differently from standard summaries, and what governance the organisation is expected to run on prompts its own clinicians write.

The company states a Responsible AI mission and says it provides source transparency along with training for ethical use, which is a real stated posture and more than much of this category offers. What is missing is everything that would let a buyer act on it.

No model or model family is named, no accuracy figure is published, no omission or false negative measure exists, no model card or evaluation methodology was located, and the granularity of the source transparency is not described, so it is not possible to tell whether a clinician can trace an individual sentence to a source document or only see a list of documents consulted. That granularity is the whole question in this category. Ask to see it demonstrated on a long and messy chart rather than on a demo patient.

Nothing identifies any party in the chain and the one document that appears to address the question does not. The company publishes a policy titled as a privacy policy, and its subject matter is contact form submissions, cookies, analytics, device and browsing data, and marketing communications, with protected health information not mentioned anywhere in it.

There is no product privacy addendum, no published processing terms, no sub processor list and no security page, and the footer carries that policy and a cookie policy and nothing else. The distinction is the finding: the presence of a privacy policy is not evidence of a position on clinical data, and a buyer scanning a footer would reasonably conclude otherwise, which is why this matters more than an outright absence would.

The policy's own security clause promises industry standard measures, which names no control and in any case governs enquiry data rather than the record. The gap is unusually wide because of what the platform touches. It reads effectively everything in the chart including outside records retrieved through health information exchanges, and both the summariser and the embedded chat assistant are generative and run on prompts the clinician writes at the point of use, so free text a clinician composes about a specific patient enters a generative layer with no stated handling at all. Get retention, training use and de identification in writing, and ask specifically whether content from one customer's records can inform outputs served to another.

Genuinely third party measured, which is rare here, and measuring the wrong thing for the risk the product carries. KLAS Research found a 31.7 point increase in Net EHR Experience Score at University of Iowa Health Care following integration, through the Arch Collaborative, a large standardised benchmarking programme rather than a vendor survey. Allina Health reports a 6x return through risk adjustment and revenue capture.

UNC Health selected the product for enterprise deployment in May 2026. An Offcall industry report ranked it the second most physician requested tool of 2025, ahead of any other chart summarisation tool, which is a demand signal rather than an outcome. A long roster of named clinicians at named institutions appear on the record with real titles, which is more accountable than the anonymous testimonial norm.

Held at B because Net EHR Experience Score measures clinician satisfaction with the electronic record, not the accuracy, completeness or safety of a summary; because attributing a system wide score move to one tool is a strong causal claim from a before and after comparison with no control described; and because the return figure is vendor reported. Nothing published measures whether the summaries were correct.

No retention period, no statement on whether customer data is used to train or improve models, and no de identification posture was located. A second pass confirms the earlier finding and sharpens it considerably.

The company publishes a document titled Privacy Policy, last updated February 2025, and it is a marketing website policy. Its subject matter is contact form submissions, cookies, Google Analytics, device and browsing data, and marketing communications. Protected health information is not mentioned in it anywhere. There is no product privacy addendum, no published data processing terms, no subprocessor list and no security page. The site footer carries a privacy policy and a cookie policy and nothing else.

That distinction is the finding on this row. The presence of a privacy policy is not evidence of a position on clinical data, and a buyer scanning a footer would reasonably conclude otherwise. The policy's own security clause promises industry standard measures, which names no control and in any case governs enquiry data rather than the record.

The gap is unusually wide here because of what the platform touches. It reads effectively everything in the chart including outside records retrieved through health information exchanges, and both the summariser and the embedded chat assistant are generative and run on prompts the clinician writes at the point of use. Retention, training use and de identification are therefore all live questions and none is answered publicly. Get all three in writing, and ask specifically whether content drawn from one customer's records can inform outputs served to another.

HIPAA compliance is explicitly claimed, with the company stating the platform was built from the ground up around HIPAA and SOC 2 Type II compliance. No business associate agreement terms are published and no statement addresses whether a BAA is included, negotiated or separately priced. That is the standard middle rung on this axis: compliance asserted, instrument unpublished. The benchmark to compare it against is a vendor that publishes the BAA text itself.

SOC 2 Type II is claimed for the company itself rather than for its hosting provider or its suppliers, which puts it on the right side of the supplier certification watchlist this index maintains, and it specifies Type II rather than leaving the report type ambiguous, which answers the standing question this index asks of every SOC 2 claim. Held at B because the claim exists only as a line of marketing copy on the homepage.

No trust centre, no security page, no status page and no independently checkable artifact was located, so a buyer cannot verify the report exists or read its scope without asking. Minor untidiness worth noting without weighting: the same section of the site renders the claim three different ways, and SOC 2 is an attestation rather than a certification.

No FDA clearance, device authorisation, regulatory pathway statement or clinical decision support exemption analysis was located. Graded C on that absence.

Two capabilities make the question live rather than academic and should be raised in diligence. The product performs patient safety indicator and pediatric quality indicator review and registry abstraction, which feed regulatory and quality reporting rather than only clinician convenience. And the embedded chat answers open clinical questions at the point of care. Ask which framework the company believes governs each surface, and to see the analysis written down.

The grade describes incentive structure and disclosure, and the credit belongs alongside it. Evidently states a Responsible AI mission covering source transparency and user training for ethical use, which is a considered stance rather than a slogan, and its clinician facing framing leads on wellbeing and cognitive burden. Against that, the commercial gradient is substantial and explicit.

The product suite includes clinical documentation improvement review, HCC review, case mix index improvement and denial appeal drafting, the site states that side effects include better quality and reimbursement, and the flagship customer outcome is a 6x return attributed to risk adjustment and revenue capture.

Risk adjustment and severity capture are among the most litigated areas of United States healthcare compliance, and drafting appeals against payer denials puts the vendor's output directly into an adversarial financial process. Case mix index makes this a second instance of documentation moving the payment unit at the admission level. Separately, no fairness, subgroup or demographic performance disclosure of any kind was located.

The stated posture is real and more than much of this category offers: a responsible artificial intelligence mission, source transparency presented as a product property, and training for ethical use, which is unusual because it treats the clinician's competence with the tool as part of the safety case rather than assuming it. What is missing is everything that would let a buyer act on it, and one gap decides whether the headline property does any work at all.

The granularity of the source transparency is not described, so it is impossible to tell whether a clinician can trace an individual sentence back to the document that supports it, or only see a list of documents the system consulted. That difference is the whole question in this category.

A document list tells a reader what was read and lets a fabricated sentence pass unchallenged, because everything in the summary points at the same pile; sentence level linkage is what makes an unsupported claim visible, and this index has credited a competitor for exactly that granularity.

No model or model family is named, no accuracy figure is published, no omission or false negative measure exists, no model card or evaluation methodology was located, and no warranty, indemnity or remediation commitment attaches. Ask to see the source transparency demonstrated on a long and messy real chart rather than a demonstration patient, and establish the granularity before relying on it.

The right shape with a naming gap. The company states it integrates with any SMART on FHIR enabled EHR and can be customised to other systems, and an integration manager customer quote describes the fastest implementation in a 23 year career, which supports the deploy in weeks claim. Ingestion breadth is a real strength: notes, labs, imaging, faxes, scanned documents and outside records. Held at B because no EHR vendor is named in text anywhere on the site.

The one system specific capability named is Care Everywhere, which is Epic's exchange, so Epic is evidently the primary target even though the marketing claims any SMART on FHIR system. An organisation outside that install base should confirm its own EHR directly rather than rely on the any EHR phrasing.

No hosting model, cloud provider, region, residency commitment or deployment option was located on any retrieved surface, including the marketing site, the case studies, the press announcements and a customer's own published documentation. Two differently phrased searches were run against the platform and its named components.

One near answer should be separated from the real one. The company positions explicitly against being a wrapper, stating that off the shelf language models are not built for the complexity of patient record data and that its platform rests on more than fifteen years of research. That speaks to model provenance. It does not say where inference runs. A proprietary model hosted in a third party environment still carries record content beyond the health system on every query, so the two questions are independent and the marketing answers only the first.

Scope makes the unanswered question material. University of Iowa's own Epic documentation confirms the platform reads the electronic medical record, Care Everywhere source data and scanned documents held in the Epic media tab, so on every summary the content of records originating at organisations other than the customer is processed somewhere unstated. Ask where inference runs under each available deployment option, whether any customer controlled option exists at all, in which region data is processed and retained, and whether the answer differs between the summarisation engine and the embedded chat assistant.

No price, tier, mechanism or contracting model is published, and every path ends at a demo request or a contact form.

The only commercial signal available is a customer reported 6x return, which describes claimed value rather than cost and should not be read as pricing transparency.

The broadest coverage in this category and it rests on instrument level work rather than on a claim of covering everything. Four distinct care settings are addressed with purpose built workflows: inpatient, outpatient value based care, emergency medicine and perioperative.

Within perioperative alone it performs pre operative review, transplant review, registry abstraction and patient safety and pediatric quality indicator review, which are specific reporting instruments rather than generic summarisation, and nothing else in this category performs them.

Named clinician users span pulmonology, nephrology, pediatric nephrology, pediatrics, urology, transplant surgery, infectious disease, hospital medicine and primary care, and the product is separately positioned for CDI specialists, denial specialists, population health teams, research teams, pharmacists and nurses rather than for physicians alone.