Wellsheet

The grade describes the mechanism, not the quality, and the product is a good one. Large language models genuinely drive the core modules: chart summarisation, generated documentation and the pathway agent. But a substantial part of what a buyer is purchasing is not the model. It is an EHR agnostic integration layer, listing in the Oracle, Epic and athenahealth marketplaces, a prioritised workflow interface that predates the generative features, and licensed third party clinical content.

The product's own lineage makes this plain: it was sold as a Smart EHR UI, a predictive workflow platform, before the summarisation layer arrived. The Elsevier acquisition sharpens the point rather than softening it, since the strategic logic is explicitly the pairing of an evidence content library with an EHR embedded delivery surface. Compare a pure summarisation engine, which has nothing to sell if the model is removed.

A wide generative and decision shaping surface with no described gate. The product generates hospital course narratives, assessment and plan text and discharge summaries, answers clinical questions conversationally in the EHR, and delivers what the company calls the AI's chosen answer when walking a clinical pathway. No confidence signal, routing threshold, abstention behaviour or required review step was located for any of it. The pathway feature does present supporting chart evidence next to its answer, which is a real verification affordance and the reason this is not lower.

One risk specific to this product deserves naming, because it is invisible in a demo. The contextual views adapt to the clinician's specialty, role and usage patterns, which means the system personalises what it shows based on what that clinician has looked at before. A prioritisation layer decides what is surfaced and therefore also decides what is not, and personalising that on prior behaviour can quietly harden an individual clinician's blind spot rather than correcting it.

Ask what the adaptation is trained on, whether anything is ever suppressed entirely rather than ranked lower, and whether the clinician can see what was deprioritised.

The company states that large language models perform the summarisation, which is more than many admit, and the pathway feature presents supporting evidence from the patient's chart alongside the answer it produced, which is a genuine output level transparency feature. Against that, no model or model family is named, no accuracy figure is published, no omission or false negative measure exists, and no evaluation methodology was located.

One structural problem compounds all of it: the site's own footer states that certain described features are illustrative of potential capabilities and not currently in the product, without saying which. That makes every capability claim on the marketing surface unverifiable from outside, including the transparency features themselves. Ask for a demonstration on a real long chart and confirm which described behaviours ship today.

Nothing identifies any party in the chain: no model or model family, no hosting arrangement and no sub processor list was located, and no retention period, training position or de identification posture was found. The one privacy artefact is weak in an unusual way, since the policy was published as a document on a consumer file sharing link rather than as a maintained page on the company's own domain, which gives it no version, no change history and no assurance that the copy a buyer reads is current.

The acquisition changes what this axis is asking, and a buyer should treat any pre acquisition position as provisional rather than settled. The company is being integrated into a global information and analytics group whose core business is publishing and data products. That is not a criticism, and a parent of that kind brings mature information governance and an evidence base that is a genuine asset to the combined product.

It does mean the separation question now exists where it did not before: the combined offering merges patient records from a health system with a licensed content platform, so ask whether patient data remains segregated from the content and analytics side of the group, whether any of it informs product development beyond the customer it came from, and whether that answer is contractual or merely current practice. Ask also for the retention schedule and a versioned policy on the company's own domain, which a parent of this size should find straightforward.

Real deployment at real scale with third party attention, and claims that do not reconcile. On the credit side: Ascension has deployed the product system wide, corroborated by Ascension's own newsroom rather than only by the vendor, and KLAS Research published a report specifically on that deployment. Named use also at Baptist, Indiana University Health, Robert Wood Johnson, Concord and San Juan Regional Medical Center.

Applying the standing precision, this is a KLAS report, not Best in KLAS, and it is a case study of one customer rather than a comparative ranking. Held at B for three reasons. The efficiency claims are unreconciled across the company's own surfaces: charting time cut by 50 percent, time in the EHR cut by 40 percent, and two hours saved per clinician daily are three different figures for adjacent claims and none carries a stated denominator or method.

The user counts are also unreconciled, with over 30,000 clinicians in earlier material, over 65,000 in a 2026 release and 10,000 plus on the current homepage. And the substantive evidence is gated: the KLAS report needs KLAS access and the vendor white paper sits behind a lead capture form, so a buyer cannot read either without identifying themselves.

No retention period, training use statement or de identification posture was located in a second pass.

The acquisition confirmed in June 2026 changes what this axis is asking, and a buyer should treat the pre acquisition position as provisional rather than settled. The company is being integrated into a global information and analytics group whose core business is publishing and data products. That is not a criticism; a parent of that kind brings mature information governance and the combined product's evidence base is its own asset. It does mean the question of what happens to clinical data now has a dimension it did not have when the company was independent.

The specific thing to establish is separation. The combined offering merges patient records from a health system with a licensed content platform. Ask whether patient data remains segregated from the content and analytics side of the group, whether any of it informs product development beyond the customer it came from, and whether the answer is contractual or merely current practice.

The training question is unanswered in either direction and should be settled in the new agreement rather than the old one.

One observation from the earlier assessment is worth keeping and is now more pointed. The privacy policy was published as a document on a consumer file sharing link rather than as a maintained page on the company's own domain. That was a weak artefact for a product deployed across a health system at scale; under a large publishing parent with established policy infrastructure it should be straightforward to replace, and a buyer can reasonably expect a versioned policy.

Ask for the retention schedule, the separation commitment, and the current policy location.

No compliance statement or business associate agreement terms were located. Agreements plainly exist, since the product is deployed across a large number of hospital sites, and the grade records that none is disclosed.

The question the earlier assessment raised is now live rather than anticipatory. The acquisition completed in June 2026, so every existing customer signed with a company that no longer exists as an independent entity.

Change of control is a real contractual event for a business associate agreement, not a formality. Establish which legal entity is now the business associate, whether existing agreements were assigned or novated, whether the parent or an affiliate is a party, and whether any customer had a consent right on assignment that was exercised or waived. Some agreements permit assignment on notice, some require consent, and a health system that has not checked may not know which it holds.

Two further questions follow from the shape of the acquirer.

The parent is part of a global group, so establish whether any entity outside the United States is now in the processing chain, and if so how that is papered. Offshore access is permitted and must be documented.

And the combined product merges the acquired platform with the parent's existing content service, which is separately deployed in hospitals worldwide. Where two products become one, a buyer should establish whether their agreement covers the combined offering or only the component they originally licensed, and whether adopting the integrated version requires new terms.

Ask for the current agreement, the assignment position, and the entity list.

A SOC 2 badge appears on the site and the certification is claimed for the company itself rather than for a hosting provider or supplier, which is the right subject and puts it above the supplier certification watchlist. Graded C rather than higher because the report type is not stated anywhere.

Type I versus Type II is the entire assurance question, since one describes controls as designed at a point in time and the other tests them over a period, and this index asks that of every SOC 2 claim. No trust centre, no security page, no status page and no scope statement was located, so nothing here is independently checkable. Ask which report exists, over what period, and to see it under NDA.

No clearance, device authorisation or exemption analysis was located. The three capabilities the earlier assessment identified all remain live, and the acquisition both strengthens the exemption argument and extends the surface it has to cover.

The strengthening first, because it is real. The combined offering grounds answers in the parent's licensed evidence base drawn from peer reviewed journals and professional bodies, and both companies state that every response is traceable to its source. Source traceability is the condition the conditional exemption for decision support turns on: it exists so a professional can review the basis rather than rely primarily on the output. A product that cites identified published evidence and links to it is doing the thing the exemption asks for, and doing it better than a system answering from an opaque model.

That argument does not resolve the classification question by itself, and the three original capabilities are where it has to be tested. A pathway feature delivering a chosen answer on a treatment question, lab interpretation, and alerts flagging patients as due for discharge each sit at a different distance from the exemption's conditions. Discharge readiness in particular is a judgement with a direct safety consequence when wrong.

The stated roadmap extends further, naming decision support, care pathways, patient engagement and analytics. Patient engagement is the one to watch, because content reaching patients directly falls outside the reasoning entirely: a patient cannot independently review the basis the way a professional can.

Ask for the written analysis, per capability, and ask whether the acquirer has performed its own before integrating.

Two genuine positives sit alongside one structural concern. First, the clinical guidance is grounded in licensed, editorially maintained third party content rather than in model inference, with UpToDate pathways, calculators and lab interpretations pre filled from the chart, and the pathway output presents supporting chart evidence.

That is the guideline grounding property this index credited in Avo: a clinician can trace the recommendation to a citable source and disagree with it on the evidence. A caution that follows and generalises: grounding the RECOMMENDATION does not ground the INPUTS. A validated calculator pre filled with a value the model extracted incorrectly returns an authoritative looking wrong answer, and the guideline provides no protection against that.

Ask how extracted values are verified before they populate a calculator. Second and separately, this record introduces a gradient distinct from the coding gradient this index usually tracks. The discharge module benchmarks against geometric mean length of stay and raises alerts for patients due for discharge, and the company's stated purpose includes ensuring no discharge is delayed. That applies financial pressure to a clinical timing decision.

It is not coding intensity and should not be conflated with it, but premature discharge carries readmission and safety consequences and no counterweight measure, readmission rate or otherwise, was located. Separately again, no fairness, subgroup or demographic performance disclosure of any kind was found.

One footer disclaimer undoes the entire marketing surface, and it is the finding on this record. The site states that certain described features are illustrative of potential capabilities and not currently in the product, without saying which.

That single sentence makes every capability claim on the surface unverifiable from outside, including the transparency features that would otherwise earn credit here: a reader cannot tell whether the supporting evidence display, the pathway behaviour or any other described property ships today or is aspirational, and a vendor that reserves the right to describe unbuilt features without marking them has made its own material unusable as evidence.

This index grades what a buyer can establish, and here the answer is nothing, by the company's own construction. What would otherwise sit on the credit side is real. The company states that large language models perform the summarisation, which is more than many admit and tells a reviewer which outputs can be fabricated.

The pathway feature is described as presenting supporting evidence from the patient's chart alongside the answer, which is a genuine output level transparency property where it exists. Neither can be relied on without confirmation. No model or model family is named, no accuracy figure, omission measure, evaluation methodology or warranty, indemnity or remediation commitment was located. Ask for a demonstration on a real long chart and a written list of which described behaviours ship today.

The strongest EHR distribution position in this category and it is verifiable rather than asserted. Three major systems are named explicitly, Oracle Cerner, Epic and athenahealth, and the product is listed in all three vendor storefronts: the Oracle Cloud Marketplace, the Epic Showroom and the athenahealth Marketplace. Being carried in a competitor EHR's own catalogue is a form of external validation, since each vendor applies its own technical review before listing.

The company describes a deliberately EHR agnostic stack built so health system IT teams can deploy enterprise wide quickly, and a system wide rollout at Ascension demonstrates that at genuine scale rather than in a pilot. The product layers on top of the record and complements native functionality rather than requiring replacement.

No hosting model, cloud provider, region, residency commitment or customer hosted option was located, and nothing describes where chart data is processed. Marketplace availability on a major cloud implies cloud delivery without establishing where data resides or who controls it.

The acquisition makes this the axis most likely to change, and a buyer should ask now rather than accept the pre acquisition arrangement as durable. Integration into a larger group routinely involves migrating infrastructure onto the parent's platforms, consolidating onto its cloud accounts, and adopting its subprocessor set. Any of those would alter the answer to every question on this row, and none of them requires telling a customer unless the contract says so.

So the questions are about the transition rather than the current state. Where is data processed today, is a migration planned, what notice will customers receive, and is the current residency position contractually fixed or merely how things happen to be configured.

The parent's global footprint makes the second part of that more than theoretical. A group operating worldwide has infrastructure and personnel outside the United States, and consolidation onto shared services can move processing or support access across a border without any product change a customer would notice.

One further question follows from the combined product. The parent's evidence service is deployed in hospitals internationally. Where two products merge, establish whether the merged service runs where the acquired platform ran or where the parent's did.

Ask for the hosting region, whether it can be fixed by contract, the subprocessor list, and the migration plan.

No price, tier or pricing mechanism is published and every route terminates in a contact form.

Two related observations. The supporting evidence is also gated, with the vendor white paper behind a lead capture form and the KLAS report behind KLAS access, so the material a buyer would use to justify the purchase is not openly readable either. And a footer disclaimer stating that some described features are illustrative rather than shipping means the scope of what is being priced cannot be established from public materials at all. Establish which features are live in the contracted version before discussing cost.

Inpatient acute care is the centre of gravity and the role coverage is genuinely broad within it. The product supports multidisciplinary rounding across physicians of all specialties, nursing and case management, and covers hospital course documentation, assessment and plan, discharge summaries and discharge planning. Contextual views adapt by specialty and by role rather than presenting one layout to everyone. Ambulatory clinics are also served.

Graded B rather than A because the depth is in inpatient workflow and care team coordination rather than in specialty specific instruments: nothing equivalent to registry abstraction, safety indicator review or a specialty coding schedule was located.