Clinical Summarization & Chart Review

Credo Health, based in Denver and led by founder and chief executive Carm Huntress, addresses the problem underneath every other product in this category: you cannot summarise a record you do not have. Its PreDx product delivers a concise clinical history, pre encounter risk analysis and HEDIS gap closure before a visit, and the retrieval machinery beneath it is what distinguishes the company. Care Map, launched in October 2025, maps a patient's complete encounter footprint. It reviews the records already held, identifies every location where care was likely delivered across hospitals, clinics and specialists both in and out of network, and crucially spots signals that another record should exist at all, the worked example being a cardiology referral with no corresponding cardiology note. It then checks each likely site against Carequality, CommonWell, eHealth Exchange and regional health information exchanges to establish what is available and what is still missing. What happens next is the part nobody else does. The company states that 30 to 50 percent of high value encounters are not returned digitally by those networks, and it pursues them anyway: AI agent assisted workflows contact source providers by telephone, through secure portals and by fax to recover what the exchanges did not deliver. Everything is then cleaned, deduplicated and stitched into a single source cited history. Buyers are value based care providers, management services organisations and health plans, with Medicare populations called out as the setting where care is most distributed and retrieval hardest. The company raised 5.25 million dollars in seed funding led by FCA Venture Partners with Hannah Grey VC, FirstMile Ventures and SpringTime Ventures, and has announced partnerships with Vim for point of care EHR delivery and with HealthMark Group on release of information. Name collision worth noting: credo.health carries content for an unrelated Digital Health Passport product. This record concerns credohealth.com.

Patients.app applies record review to two narrow, high consequence workflows rather than to general chart summarisation, and both are instrument level rather than generic. The perioperative product screens records for surgical readiness. It retrieves patient data from outside the organisation, including from referring sites, reviews it against the surgery centre's own criteria, surfaces what is complete and flags what is missing, and auto generates intake and anaesthesia reports so nursing review is faster. Missing laboratory results, absent clearances and surgical risks are flagged before they delay or cancel a case, and the company reports 30 to 60 minutes saved per case. Every result links back to its source. A secondary commercial argument is that better workups capture missed codes and reduce denials. The transplant product is the more consequential one. It scans thousands of pages across EMRs, PDFs and scanned documents to flag key criteria, surface contraindications and highlight strong candidates against the transplant centre's own guidelines, generates review committee reports, and fills UNOS and TIEDI forms. It also guides patients through the process with support from trained transplant recipients, a human peer layer nothing else in this index has, and the company states clinical oversight is provided by a named physician. That combination makes this the sharpest governance record in the lane, and the reason is set out in full on the governance axis rather than here. A system that highlights strong transplant candidates is operating adjacent to decisions about access to a scarce, life saving resource, in a domain where disparities in access are among the best documented inequities in American medicine. Nothing published describes how the model performs across populations. No customer is named, no funding was located and no outcome data is published, so several axes below are Not Rated.

SCOPING NOTE FIRST. Vibrant Practice is not a summarisation tool, it is an AI native electronic health record and practice operating system, and it serves a market nothing else in this index covers: independent clinicians practising functional, integrative and longevity medicine, in concierge, direct primary care and hybrid models across North America and Europe. It is filed here because the clinical AI work it does is synthesis, reading a patient's data and laboratory results and turning them into charting and treatment plans, and cross listed into administrative automation for the intake, scheduling and billing side. The platform automates intake, documentation, appointment scheduling and billing, generates personalised treatment plans from patient data, lifestyle factors and medical history, and includes a patient application providing continuous communication, health tracking and personalised recommendations. Its clinical pitch centres on laboratory work, which is the right emphasis for its market: functional and longevity practice runs on large panels that a clinician must interpret and trend, and the company markets simplified labs, reduced charting time and faster clinical insight. It was founded in 2024 in Sacramento by Sunita Mohanty and Pedro Tabio. Mohanty was previously an AI product leader at Meta working on the Ray-Ban smart glasses and Oculus platforms, holds a Stanford Graduate School of Business degree, and speaks at the Institute for Functional Medicine and A4M. She has publicly argued that human in the loop is the correct approach both for coding and for anything relational or patient facing. The company raised 1.7 million dollars in pre seed funding in March 2025 from Anthemis Group, Dria Ventures, Emerson Collective, Hustle Fund and Lombardstreet Ventures, and sells on subscription. No customer is named, no deployment figure is published, and nothing about models, accuracy or clinical validation was located, so most axes below are Not Rated.

aiomics is a Berlin company building clinical data intelligence for European rehabilitation and acute care hospitals, founded by Sven Jungmann, a physician who describes watching paperwork erode time and purpose at the bedside. It raised a 2 million euro pre seed round in September 2025 led by Vorwerk Ventures with Calm/Storm, Norrsken Evolve and Rule30 participating. The product reads what a hospital already has, in whatever form it arrives. Speech, handwriting and existing documents, including PDFs, scanned letters and printed reports, are captured into a single validated workflow, structured, checked for completeness and consistency through a human in the loop process, and returned to existing systems as structured, coded, audit secure data with FHIR based handoff. From that structured record it drafts admission notes, discharge summaries, therapy briefings, insurer packets, rehabilitation applications and payer correspondence, all for physician review and approval, with interactive checklists ensuring packets are finished rather than left with gaps. The distinguishing feature is what it looks for rather than what it writes. aiomics surfaces contradictions, gaps and missing detail in the record itself, which is the failure mode this category is defined by and which almost every competitor treats as somebody else's problem. A physician on the company's own site puts it more sharply than any vendor copy does: for the first time someone understands how incomplete the records really are and does something about it, instead of copying the errors forward. Its compliance posture is built for Europe rather than adapted to it. All processing remains in the EU, infrastructure is C5 attested, information security is externally audited, and the company holds ISO/IEC 27001:2022 certification issued by TÜV NORD, while working toward ISO 13485 and Medical Device Regulation alignment and referencing the EU AI Act directly. Most notably for this index, the effectiveness of the product is currently under independent scientific evaluation at the Charité Institute for Medical Informatics, with results to be published on completion. The company states plainly that it believes in evidence over claims, for its own tools as much as for a hospital's records, and it caveats its own efficiency figures by noting that actual time savings depend on individual documentation.

MedScrub inverts the usual arrangement in this category. Most products here send patient records to a model. MedScrub strips the patient out of the records first, sends the de identified content to whichever model the customer chooses, and re identifies the answer on the way back. All eighteen HIPAA identifiers are removed by a proxy before data reaches any model, the substitution is reversible, and the company reports 78 percent token cost savings as a side effect of the compression involved. That proxy is sold on its own as an HTTP API and as an MCP server, with self hosted or cloud deployment, and it is deliberately model agnostic: GPT-4, Claude, Gemini, Mistral and local models running under Ollama are all named as supported targets, on the argument that once identifiers are stripped any model becomes safe to use. Developer use cases named alongside the clinical ones include intake and triage chatbots, ICD-10, CPT and SNOMED coding assistance, referral routing and care plan generation, and using de identified production data in staging environments. On top of that sits a clinician facing product. It pulls conditions, laboratory results, vitals, medications and imaging from the EHR and produces a visit summary before the patient arrives, and it answers questions about any chart with the same de identify and re identify path in between. The company is explicit that this starts from structured EHR data rather than from a recorded conversation, positioning it against ambient capture. One architectural claim deserves particular attention because nothing else in this category makes it: patient data syncs from the EHR into a clinical data repository the customer owns rather than into the vendor's cloud. Combined with self hosting and local model support, that means an organisation can run the whole path without patient content leaving infrastructure it controls. Note on identity: this record is built on the product, MedScrub at medscrub.ai. No corporate parent was independently verified and none is asserted here.

ThetaRho sells a clinical intelligence platform in three deliberately separated layers, and it will sell you any of them. A normalisation layer converts raw data from athenahealth, Epic, Cerner, Meditech, health exchanges reached through CommonWell and TEFCA, wearable and remote monitoring devices and public research corpora including PubMed into clean FHIR R4 resources, deduplicated and reconciled into a single view and annotated with RxNorm, LOINC, SNOMED CT, ICD-10 and MeSH. An AI layer adds semantic embeddings and clinical natural language processing so that a query for blood pressure medications returns ACE inhibitors, ARBs and calcium channel blockers even where those words never appear in the record. An application layer sits on top, and the company states that any EHR vendor, health system or clinical AI company can build on the first two layers rather than rebuilding normalisation themselves. Two applications are its own. RISA retrieves a patient's longitudinal record when a physician opens the chart and surfaces only what matters for that visit. DataDoc is a natural language chat window embedded directly in the athenahealth patient chart, letting a clinician ask for medications, labs, history, conditions and outside records rather than clicking to find them. Both are carried on the athenahealth Marketplace, with RISA described as certified there. The company is unusually direct about its own framing, arguing that clinical AI which knows medicine but not the patient is really just search, and positioning its work as the context retrieval that precedes physician judgement rather than as a substitute for it. It also does something almost nothing else in this index does: it names the models it runs on, citing Llama, Aloe and GPT models for clinical question answering grounded in patient data with traceable citations. That candour is credited on the transparency axis. The same sentence, however, ends with the claim that nothing is hallucinated, which is an absolute assertion of a kind this index treats as unfalsifiable, and both halves are recorded.

Fourier Health, based in Miami and founded in late 2023, automates medical record intake and turns it into use case specific clinical summaries. It ingests PDFs, faxes, handwritten notes, EHR data and records retrieved from health information exchanges, classifies, labels and structures them, and produces summaries shaped to the workflow they are for, so a referral summary and a pre visit readiness summary differ even when built from the same records. Delivery is deliberately flexible: embedded in the EHR, pushed to an inbox, exposed via API, or through a standalone web application, as structured fields or as a PDF. The founding team is unusual and relevant. James Lloyd, the chief technology officer, previously co founded and served as CTO of Redox, the healthcare interoperability platform integrated with thousands of healthcare systems. Christopher Lee, the chief executive, previously co founded and served as chief operating officer of InfiniteMD, an expert medical opinion platform acquired in 2021. The company raised 8.4 million dollars in seed funding in October 2025 led by Yosemite, with Innospark Ventures, NextGen Venture Partners and Tau Ventures participating, following pre seed backing from Lasagna, NextGen, Myelin and Despierta. Two design choices define it. Every summarised element is attributed back to its location in the original records, which the company frames as accountability and audit proofing rather than as a convenience. And a proprietary network of specialist clinicians reviews and validates summaries, so the human layer is verification rather than production. Deployment is described as dozens of provider organisations, none named, and the product is listed on the AVIA marketplace. One thing a buyer should weigh. The company positions the work inside the revenue cycle, describing the process as reimbursable and framing the benefit as improving clinical decision making while increasing revenue through better risk capture and coding accuracy. That is a commercial argument as much as a clinical one, and it is discussed on the governance axis.

Apricot, based in Oklahoma City, builds AI documentation for home health and post acute care, and it is the fifth vendor in this index addressing that segment alongside Lime Health, Enzo Health, Andy AI and Roger Healthcare. Founder and chief executive Trent Smith ran his own home health and hospice agency for seven years before starting the company, and the product reflects that: it began with the Start of Care visit, the single most demanding documentation event in home health, where a nurse works through hundreds of required OASIS fields while the patient waits. Its architecture is deliberately different from every other home health vendor here and the company argues the difference openly. It is not ambient. Before a visit the clinician sees the referral document and attachments with an AI summary highlighting relevant history and areas of concern. During the visit the product captures structured content, time in home, consents, vital signs, assessments, photographs, notes and targeted clinician controlled audio, with an in home guide to prevent omissions. After the visit the clinician completes a guided interview and the documentation is drafted from that. The company positions this explicitly against what it characterises as riskier ambient approaches, and its own representatives state that it drafts OASIS forms from patient documents but does not replace clinical decisions, with transparent reasoning and agency review leaving final decisions with the team. Coverage of the Start of Care is described as complete, spanning medication reconciliation, wound documentation, calendar plotting, interventions and goals and narratives, with support extending to other OASIS, therapy and routine visits. It integrates with Netsmart myUnity, the post acute care record, and appears on the Netsmart marketplace with a co branded page, and it is also demonstrated for Axxess users. The company raised a Series A led exclusively by Insight Partners in October 2025, that investor's first investment in an Oklahoma company, with Cortado Ventures also on the register. Customers are described as ranging from regional operators to national leaders but none is named.

DELIBERATELY THIN RECORD. Galen Health occupies a segment nothing else in this index covers, which is why it is here, but almost nothing about the company is publicly verifiable and most axes below are Not Rated for that reason rather than because the posture is poor. Galen sells what it calls AI teammates for tumor boards. The multidisciplinary tumor board is the forum where a cancer patient's case is presented to surgeons, medical and radiation oncologists, pathologists and radiologists together, and a treatment recommendation is agreed. Preparing those cases is heavy manual work and the meetings themselves are time constrained. Galen addresses the whole cycle: preparing the case presentation beforehand, identifying candidate clinical trials, retrieving patient information and insights live during the meeting, and maintaining a knowledge base of the care decisions the board reaches so that past discussions can inform later ones. The company positions it as a single layer unifying oncology workflows across multiple hospitals and clinics. That combination is genuinely distinctive. This index already covers clinical trial matching as its own category and chart summarisation as this one, but nothing else addresses the tumor board as a workflow, and no other product reviewed here operates inside a live multidisciplinary meeting rather than at an individual clinician's desk. What is missing is everything a buyer would use to evaluate it. No customer is named, no funding is disclosed, no deployment is documented, no model or accuracy information exists, and the efficiency claims, hours saved weekly and millions in hospital cost reduction, carry no substantiation. The company describes itself as trusted by leading cancer centres without naming any. Its compliance language is HIPAA ready rather than HIPAA compliant, a distinction discussed on the relevant axis. Treat this record as a placeholder documenting a real and uncovered segment, to be refreshed when the company publishes more.

RhythmX AI, based in Palo Alto and launched in 2023, sells a precision care platform aimed at primary care physicians. It unites data across more than ten sources, EHR content alongside clinical policies and guidelines, financial, payer, social and lifestyle data, and produces patient specific recommendations covering laboratory tests, imaging, medications, follow ups, social care and referral routing, which a clinician can drill into through a generative interface. It is EHR agnostic and runs inside Epic at its named customer. The chief executive is Deepthi Bathina, previously chief clinical product officer at Humana, and the company is owned by SAI Group, the private investment firm behind SymphonyAI and ConcertAI. It is filed under clinical decision support rather than clinical summarisation, and cross listed into summarisation, because of what it produces. Chart consolidation is real and clinicians describe it as replacing extensive manual review, but the product's distinguishing output is a forward looking recommendation about what to do next rather than a report of what the record already contains. That is the boundary this index draws between the two categories. The substantive evidence is a single named deployment and it is a good one. Presbyterian Healthcare Services in New Mexico, a nine hospital system with its own statewide health plan, expanded the platform to 200 primary care clinicians in February 2026 in what the companies describe as the first full system deployment of a precision care AI platform. Within weeks of the initial pilot, clinicians identified and reviewed more than 200 combined HCC and non HCC conditions across more than 10,000 patient encounters, and the system's chief medical information officer is on record describing consolidated information at the point of care and an assistant answering patient specific questions. A clinical advisory board carries named executives from Prime Healthcare, Sentara Health and Mass General Brigham. One attribution caution. RhythmX has merged with the patient engagement company Get Well and now also appears as GW RhythmX. The frequently quoted reach of 150 health systems and 85 million patients is the COMBINED figure for both companies and should not be read as this platform's deployment base.

Autonomize AI, founded in Austin in 2022 by Ganesh Padmanabhan and Kris Nair, sells a multi agent orchestration platform for healthcare knowledge work. It describes more than 160 pre built agents and copilots that turn unstructured material, clinical notes, PDFs, faxes and claims, into structured context for a reviewer to act on. The company raised a 28 million dollar Series A in 2025 and reports deployments at Fortune 100 healthcare organisations and top 20 pharmaceutical companies. The commercial centre of gravity is the health plan, which is why this record sits in revenue cycle and prior authorization rather than in clinical summarisation. The flagship is a Prior Authorization Copilot handling inpatient, imaging, cardiology, durable medical equipment and other service categories, designed to interface with a plan's existing medical management and medical policy systems rather than replace them. Around it sit copilots for payment integrity and pre payment review, medical versus pharmacy benefit determination, HEDIS care gap analysis, case management and clinical trial planning. It is cross listed into clinical summarisation because one of those agents is a genuine chart review product. The Medical Record Review Copilot aggregates multimodal charts from multiple sources and formats into a unified searchable view for chart reviewers to summarise and analyse, and it is distributed through the Microsoft marketplaces alongside the care gap copilot. Reported outcomes are efficiency figures rather than accuracy ones: care management teams spending 78 percent less time per case, an 85 percent improvement in case review efficiency, prior authorisation moving from twenty or thirty minutes to seconds, an 80 percent reduction in manual errors, and up to 55 percent savings in clinical and non clinical staff time on prior authorisation. All are vendor reported without stated baselines or methods. The company names real time governance, explainability and a human in the loop design in which clinical teams retain control as platform level differentiators.

Retrieve Medical Holdings, based in Bedminster, New Jersey, sells two clinician facing products built on the same engine. Retrieve Dx, also marketed as Retrieve Dx/MDM, runs an intelligent search across a patient's record inside Epic, Cerner, Meditech or another EMR, reading labs, notes and imaging reports, and surfaces previous diagnoses, pre existing issues, comorbid and major comorbid conditions, abnormal results and potential risk factors to the physician for validation. PreviewMD, announced in 2025, applies the same technology to outpatient visit preparation, scanning up to a year of history over a customisable interval and producing a clinical note ready to drop into the EHR, formatted as a consult request or admission note, with optional health information exchange access for records held elsewhere. Two design properties are worth noting. Anything the software highlights can be opened in place to show the actual note or result in its original context inside the EMR, so a physician verifies against the source system rather than against the vendor's rendering. And the product writes back: once the physician decides an item is relevant, a single action pulls the underlying data and enters it into the chart in the correct format and location. Most products in this category read the record and stop; this one closes the loop. The company is chaired by Mark Rosenberg, a past president of the American College of Emergency Physicians, and is publicly traded, which is recorded here under supplier continuity rather than as a comment on product quality. A buyer should understand what the product is optimised for. It is marketed as clinical documentation integrity, and the company states its case directly: surfacing additional comorbidities raises the Case Mix Index, that index directly affects reimbursement, and hospitals can anticipate increased revenue over time. It also reports a substantial reduction in physician queries, which is a genuine burden benefit rather than a revenue one. Both belong in the assessment.

Zus Health assembles a longitudinal patient record from national data networks and makes it usable. The Zus Aggregated Profile, or ZAP, pulls EHR data from thousands of provider sites through CommonWell and Carequality with TEFCA planned, prescription fill data through Surescripts, and admission, discharge and transfer alerts through PointClickCare and Bamboo Health, then deduplicates, normalises to FHIR, applies consistent terminology and presents the result to a care team. Coverage is stated at more than 320 million patients with a roughly 90 percent success rate matching a patient to data. Identity is resolved by the company's own master patient index, which assigns a Universal Patient Identifier shared across every record believed to belong to the same person. The summarisation layer is real and clinician facing. GPS sits alongside the ZAP overview and gives a structured view of history, problems, medications and unstructured notes tailored to the care model, used by early adopters to accelerate visit preparation, and every summarised point carries a direct link back to the underlying record. Separate enrichments include hospitalisation summaries, which the company says detect 33 percent more admissions and discharges than prior vendors, and medication journey normalisation that groups prescriptions by active ingredient. It is filed under administrative automation rather than clinical summarisation, and cross listed into summarisation, because of what the deliverable is. For a product like Abstractive Health the retrieval exists to feed the summary. Here the summary is one of several ways to consume the data, alongside REST and GraphQL APIs, webhooks and analytics tables pushed into Snowflake, BigQuery or Databricks. Buyers purchase the network and the normalised record; the summary makes it readable. That places it alongside Particle Health rather than alongside the clinician facing summarisers. One architectural detail is worth understanding before diligence. Access to third party data is gated on the customer asserting an active treatment relationship, and where that assertion is absent an organisation sees only the data it contributed itself. That is an access control tied to a legal basis, which is more than most platforms describe, and it rests on customer self attestation.

emtelligent, based in Vancouver, British Columbia, sells the extraction layer rather than a finished clinician facing application. Its Medical Language Engine turns unstructured clinical text into structured data mapped to clinical ontologies, Document Manager splits, digitises and collates bundled PDFs, paper forms and complex medical documents, and Clinical Workflow is an AI assisted review interface for clinicians, coders and reviewers with free text and code based search across ICD-10 and SNOMED. Its health system offering produces summaries so care teams can see the whole record rather than the portion their own system holds. Cofounder and chief executive Tim O'Connell is a practising physician; the chief technology officer and cofounder is Anoop Sarkar. The engine is unusually specific about what it actually does, and the named capabilities are the ones that matter most in this category. Alongside entity linking and ontology mapping it performs polarity and uncertainty detection, distinguishing an asserted finding from a negated or hedged one, and experiencer detection, distinguishing a condition the patient has from one a family member had. Those are the two classic failure modes of clinical text processing and the two most likely to corrupt a summary silently. It also handles measurement and temporality, relations, and medication identification with follow up detection. The company positions explicitly against general purpose generative AI, stating that unaligned models are not accurate enough for medical use, are prone to hallucination, and have had difficulty referencing source data reliably enough to permit proper human review. Naming non determinism and source referencing as the problems being solved is a more candid framing of the technology class than most vendors offer. Buyers span payers, health systems, pharmaceutical and life sciences companies, and health technology and data services providers. That last group is a significant distribution channel: emtelligent frequently runs as embedded infrastructure inside larger data aggregator platforms, structuring clinical text before it feeds downstream analytics, with Optum described as a beta customer. Products can be deployed in the vendor's environment, in the customer's own environment, or in a private cloud.

Wellsheet sells a Care Team Copilot that sits on top of the EHR rather than replacing it, reading the full chart to produce prioritised clinical views, narrative summaries, and generated documentation for hospital course, assessment and plan and discharge summaries. An embedded chat agent answers patient specific questions inside the EHR, and an AI Pathways feature walks evidence based pathways, presenting its chosen answer alongside the supporting evidence drawn from the patient's own chart. The company is explicit about the distinction from ambient scribes: it does not listen to the encounter, it reads what is already documented. It is inpatient and care team centred rather than physician only, covering multidisciplinary rounding across physicians, nursing and case management, and discharge planning with a dashboard tracking geometric mean and average length of stay plus alerts for patients due for discharge. Contextual views adapt to the user's specialty, role and usage patterns. Clinical content is licensed rather than generated, with UpToDate pathways, calculators and lab interpretations pre filled with patient specific data. Two pieces of corporate context matter to a buyer. Elsevier has acquired Wellsheet, positioned as closing the gap between patient data and clinical evidence at the point of care. That is a continuity positive relative to a venture funded independent, and it raises one specific question worth asking directly: the integration the product is built around is UpToDate, which belongs to Wolters Kluwer, a direct competitor of Elsevier in clinical reference content. Confirm in writing that the UpToDate integration survives, and on what terms. Second, Ascension has deployed Wellsheet system wide, confirmed by Ascension's own newsroom, alongside named use at Baptist, Indiana University Health, Robert Wood Johnson, Concord and San Juan Regional Medical Center. One caution about reading the marketing. The company's own website carries a footer disclaimer stating that certain features described are illustrative of potential capabilities and are not currently included in the product. It does not say which. Disclosing this at all is more candid than most of the market, and it also means no feature on the site can be assumed to ship. Confirm each capability against a live instance before it enters an evaluation.

Evidently, based in San Francisco, sells what it calls Clinical Data Intelligence: a layer that reads effectively everything in a patient's record, including labs, notes, imaging, scanned documents, faxes and outside records pulled through Care Everywhere and other exchanges, and turns it into summaries, drafts and answers inside the EHR. Three surfaces sit on that engine. AI Summaries produce a full chart summary on any clinical concept or a custom prompt. AI Drafts generate documentation in an editor. Ask Evidently is a conversational assistant embedded in the EHR that has already read the chart and will retrieve, summarise or draft on request, described by the company as a built in resident who has pre read everything. It is the broadest product in this category by care setting. Inpatient work covers admit notes, discharge summaries, clinical documentation improvement review and denial appeal drafting. Value based care covers HCC review and care gap reconciliation. Emergency medicine covers reading the chart on arrival, answering questions in medical decision making, note drafting and handoff. Perioperative covers pre operative review, transplant review, registry abstraction and patient safety and pediatric quality indicator review. Registry abstraction and safety indicator review are functions nothing else in this category performs. Named customers include University of Iowa Health Care, Allina Health and UNC Health, which selected Evidently in May 2026 for an enterprise deployment across its Triangle region hospitals and clinics. A study by KLAS Research found a 31.7 point increase in Net EHR Experience Score at University of Iowa Health Care after integrating the product, and Allina Health reports a 6x return through value based care risk adjustment and revenue capture. Unusually for this category, a long list of named clinicians at named institutions appear on the record by name and title rather than as anonymous quotes. One feature deserves specific attention before purchase. Because summaries can be generated from a user written custom prompt, and because the chat assistant answers open ended questions, the output surface is defined by the clinician at the point of use rather than by the vendor in advance. Whatever validation exists cannot cover a prompt the vendor never saw. That is a real and undisclosed shift of evaluation burden onto the buying organisation, and it is discussed on the autonomy axis.

Pieces is a physician led clinical AI company from Irving, Texas, founded by Ruben Amarasingham MD, who previously founded the Parkland Center for Clinical Innovation, one of the first applied clinical AI institutes embedded in a public health system, and directed biomedical informatics at UT Southwestern. Its cloud hosted Pieces Intelligence Platform condenses and summarises clinical data directly inside the EHR and pre generates progress notes, discharge summaries and multidisciplinary care plans for physicians, nurses and case managers. The company reports more than 10 million AI generated clinical documents produced to date and is backed by NIH funding. Current status matters here. Smarter Technologies acquired Pieces on 30 September 2025 and folded it into a new product, SmarterNotes, which combines the Pieces documentation workflows with SmarterDx clinical AI. Smarter Technologies was itself formed in May 2025 under New Mountain Capital from Access Healthcare, SmarterDx and Thoughtful.ai. The Pieces brand no longer has an independent web presence and piecestech.com now redirects to smartertech.com; the founder is now Chief Medical Officer of SmarterDx. Buyers should also note what the combination does to the product's purpose: SmarterNotes is marketed as producing notes optimised for reimbursement from the start, connecting admission to final payment, preventing queries and denials and identifying missed revenue opportunities. A summarisation tool has been fused to a revenue cycle engine, and that is a different product from the one Pieces sold alone. One attribution caution. The headline performance figures published alongside SmarterNotes, including 12 million cases analysed, a 5 to 1 return from day one and roughly 2.5 million dollars in annual net new revenue per 10,000 discharges, are attributed by the vendor to the SmarterDx offering, not to the Pieces summarisation product. Do not read them as evidence about the summariser. Pieces is also the subject of the first state enforcement action against a healthcare generative AI vendor in the United States, described in full on the governance and regulatory axes of this record. That matter was resolved without any monetary penalty, without any admission, and the company denies wrongdoing.

Abstractive Health retrieves a patient's full longitudinal record from the national health information exchange networks and summarises it, with every sentence in the summary linked back to the source note it came from. Founded May 2022 in New York City by Vince Hartman, Ritika Poddar and Giordana Pulpo, out of Cornell Tech and a research partnership with Weill Cornell Medicine. Hartman was previously a product manager at Epic and at Robin Healthcare. Two cofounders were named to Forbes 30 Under 30 for Healthcare. The company is a Certified B Corporation. The product is a direct implementer on Carequality and CommonWell rather than a reader of one EHR, and it performs geospatial retrieval around a patient's current and prior addresses to assemble a record from institutions the treating organisation does not share a system with. Incoming data arrives as CCDAs, FHIR R4 resources, PDFs and image based documents including handwritten scans, and is normalised into a single structure, with the original CCD and FHIR files available for download. Outputs include medical summaries, handoff notes, discharge summaries, transition of care notes and machine readable JSON, CDA and FHIR. It runs as a SMART on FHIR app, a web app or a Chrome extension, and names twelve supported EHRs. A retrieval augmented question answering module answers chart questions with citations, and a separate module issues clinical decision support alerts. The reason this record matters to the category is the evidence. In December 2024 the company's own team published a peer reviewed cohort study in JAMA Network Open, with Weill Cornell and NewYork Presbyterian co authors and institutional rather than vendor funding, evaluating its summaries against physician written notes across 1,600 emergency medicine records. On conventional automated summarisation metrics the model beat the physicians. On a purpose built clinical evaluation framework it was marginally inferior on usefulness and safety, with safety concerns arising in 6 percent of cases, the named failure modes being data omissions and mischaracterisations. The authors concluded that a physician in loop design is necessary. Almost nothing else in healthcare AI publishes a result that undercuts its own product, and buyers should weigh it accordingly. They should also note that the company markets the same study under the badge Clinically proven, which is a considerably warmer characterisation than the study's own conclusion supports.

Regard reviews the entire electronic health record and recommends diagnoses to the clinician at the point of care, then generates the note around them. Founded 2017, based in Los Angeles and New York City. The company frames this as a diagnosis problem rather than a documentation problem: its stated premise is that physicians see roughly 3 percent of the data in a chart, and its diagnostic intelligence layer reviews the rest and surfaces conditions with the supporting chart evidence attached. Four modules run off the same engine: Clinical Notes, Mid Revenue Cycle, HCC Capture and Screening. In July 2025 the company added Proactive Documentation, combining chart data with ambient conversation from the room, plus an agent named Max, and described the platform as expanding from a hospitalist tool to system wide coverage. Named health system customers include Sentara Health, WakeMed, Penn Highlands Healthcare, Kettering Health, Main Line Health, FirstHealth, Westchester Medical Center, UAMS, Eisenhower Health and Torrance Memorial. Sentara reports a 17 percent increase in CC and MCC capture alongside a 4x return per user. The company reports 12,993,284 recommended diagnoses accepted by clinicians, and publishes site level figures including 50 million dollars in revenue earned at an Arizona health system, 9.3 million dollars in denials prevented at a North Carolina system and a 20 percent reduction in queries at a Pennsylvania system. Two things a buyer should weigh. First, Regard holds a regulatory artifact that is rare in this category and independently verifiable: an ONC Health IT Module certification, number 15.04.04.3192.Rega.01.00.0.240502, certified 2 May 2024 against 2015 Cures Update criteria covering electronic health information export, authentication and related privacy and security criteria. That is neither FDA clearance nor a HIPAA attestation, and no device authorisation or published clinical decision support exemption analysis was located for a product whose core function is recommending diagnoses. Second, every headline outcome the company publishes is financial or operational. None of them measures whether the recommended diagnoses were correct, and the acceptance figure is published as a count of accepted recommendations without the number recommended, so an acceptance rate cannot be derived from it.