Navina

The core product runs on its models: Navina's core technology page describes proprietary language models that classify documents and extract codes and dates from free text, a suspected conditions engine that surfaces chronic conditions never documented before, diagnosis inferencing for HCC risk adjustment, and generative AI that drafts MEAT compliant documentation. Inferring an undocumented condition from scattered evidence is not a retrieval task, and those insights do not exist without the models.

The platform underneath carries value of its own. Navina reconciles EHR, health information exchange, claims and care gap files into one Patient Portrait, monitors screening and vaccination gaps, and sells analytics dashboards that track risk adjustment and quality performance across an organization. Remove the models and a value based care group still has data aggregation, care gap tracking and performance reporting.

Oversight is built around evidence: every insight is linked to the original clinical evidence, and diagnosis suggestions are presented at the point of care for the clinician to accept or reject, with documentation drafted for them to confirm. Customer physicians quoted on the home page cite that evidence as what makes the tool trustworthy. For a system whose suggestions raise risk scores, traceability to source is the control that separates decision support from coding pressure.

The thresholds are missing. Nothing published says what level of evidence surfaces a suspected condition, what the system may not suggest, or what happens to a suggestion a clinician does not address. The core technology page reports a 77 percent acceptance rate, which shows use, not where the boundary sits. The home page speaks of accurate diagnosis, complete documentation and capturing a complete picture of each patient's health; in each sentence the clinician is the one documenting with Navina's help, so these are not claims that the output is complete.

Architecture is described with real specificity: proprietary language models rather than a generic wrapper, document type classification, structured extraction of codes and dates from free text, and segmentation for long complex records. The company publishes an explicit position against black box models, stating insights are clearly explained and directly referenced to original clinical sources.

What is not published is model performance measurement, validation methodology, or accuracy of the suspected condition inference, which is the number that matters most given what those inferences drive.

The architecture is described and the model layer is not named. Navina describes its own proprietary language models for document classification and extraction and hundreds of proprietary algorithms, and separately says it uses state of the art generative AI to draft MEAT compliant documentation, without saying whose model that is or whether it runs outside Navina's own environment. The host is named in the privacy policy: AWS, North Virginia. Record the host for residency and breach scope; it does not answer which model sees the record.

No subprocessor list is published, and the privacy policy allows anonymized or de identified information to be disclosed to third parties without restriction, so the party list for de identified patient data is open as well.

The home page names the headline evidence: an independent study by the American Academy of Family Physicians, published as a report, found a 30 percent reduction in chart review burden and a 23 percent reduction in burnout among clinicians using Navina. That is outside evaluation of the product in use.

It answers a different question from the one the product is sold on. The headline claims are about risk adjustment and diagnosis: 500,000 previously undocumented conditions identified in the last 12 months, a 77 percent acceptance rate for the inference engine, and customer figures such as more diagnoses per visit and higher risk scores. Those are vendor reported with no method, denominator or baseline, and no published analysis separates appropriate condition capture from score inflation, which is also what CMS risk adjustment audits examine.

The privacy policy of Navina Technologies Ltd covers the personal data of its customers' end users, patients included, and answers part of the lifecycle. Data processed on behalf of a customer is retained for as long as the service runs and a reasonable time after; data is stored on AWS in North Virginia; and information the company anonymizes or de identifies may be used and disclosed to third parties without restriction, including to improve its services. The core technology page adds that Navina is built with strict data governance, HIPAA compliance and encryption.

What is missing is a direct answer on training. Nothing says whether identifiable customer records train or improve the models, and the de identified data clause permits it. The core technology page also describes the training data as diverse and representative of all demographics without saying where it came from. No retention window in days, deletion procedure or end of contract disposition is published.

A status is stated and supported by a privacy document that reaches the product. The home page carries a HIPAA badge beside its SOC 2 and ISO 27001 marks, and the core technology page says the platform is built with strict data governance and HIPAA compliance. The privacy policy of Navina Technologies Ltd covers the personal data of customers' end users, patients included, processed on behalf of customers, with retention tied to the service period and storage on AWS in North Virginia.

What is not published is the agreement itself or the words business associate: no BAA, template or summary of terms, and nothing on which tier or contract it applies to. Navina reads the patient chart on behalf of a provider organization, so an agreement would normally be in place; it is simply not visible before a sales process. Whether its terms are negotiable is not stated, nor how its deidentified data terms sit beside the privacy policy's clause allowing anonymized information to be used without restriction.

Navina presents ISO 27001 and SOC 2 as its own: the home page carries AICPA SOC 2 and ISO 27001:2022 badges, and its article on its ethical approach to clinical AI states ISO 27001 certification and SOC 2 Type II with annual evaluations by third party auditors. Named controls include least privilege and multi factor authentication.

Nothing sits behind the badges in public: no trust center, no certificate with its scope and expiry, no penetration testing statement and no subprocessor list. The assessment boundary is not described, so it is unclear which systems and entities fall inside it (including the generative documentation component) or when each audit period closed.

No clearance, authorization or regulatory position is stated. What the product does is suspect diagnoses: the suspected conditions engine surfaces chronic conditions never documented before, and diagnosis inferencing proposes HCCs, each presented with a link to the evidence it rests on (notes, imaging, labs, medications, exchange and claims data) for the clinician to confirm or reject. The core technology page says every insight is linked to the original clinical evidence, and the suggestions are sold for documentation and risk adjustment rather than as finding disease to treat.

That makes it decision support whose basis the clinician can open and read, whoever produced the document. A suggestion resting on a model inference with no document behind it would be a different matter. The evidence is not all of one kind, though. A claim shows a condition was billed, not that it is present now, and exchange data can carry a diagnosis forward for years, so a suggestion resting only on those invites documentation of something not assessed at this visit. The commercial side is structural rather than an accusation: a surfaced diagnosis that is documented also raises the risk score.

Navina publishes a stated ethical approach to clinical AI centered on transparency and clinician trust, positioned against black box models, and its core technology page says it proactively monitors for and addresses implicit biases in its models and that diverse, inclusive training data representing all demographics facilitates equitable recommendations. That is policy language with a claim of equity behind it and no result: no subgroup performance, no monitoring metric and no bias evaluation is published.

The claim matters in a specific way here. A model that surfaces suspected conditions from documentation may find more conditions in patients with richer records, systematically favoring those with better historical access to care, and the inputs make the relevant groups reportable: insurance type, documentation density, age, and how much of a patient's record arrives through exchange or claims rather than the practice's own notes.

The mechanism is real and the missing number is the one the product turns on. Insights are directly referenced to original clinical sources and the company publishes an explicit position against black box models, so a clinician or coder receiving a suggestion can follow it to the documentation that produced it and reject it on the evidence.

Stating a position against opacity, rather than merely happening to be traceable, is a commitment a customer can hold the vendor to as the product changes. What is absent is any measurement of the suspected condition inference, which is the output that matters most here.

A suspected condition surfaced to a clinician before a visit shapes what is assessed and, once documented, what is coded and therefore paid, so the two error directions have different consequences: a false suggestion invites documentation of something that was not assessed, and a missed one leaves a real condition uncaptured. No precision, recall or validation methodology is published for it, no evaluation methodology is described, and the published terms carry no warranty, indemnity or remediation commitment.

Named, in production and documented from several directions. The home page shows athenahealth, Epic and Veradigm as EHR integration partners. Navina's February 2024 announcement describes the platform live inside the Epic environment of NOMS Healthcare, with diagnosis recommendations presented in Epic's Best Practice Advisory alongside a link to the source evidence, and Navina's own listing on the Avia marketplace names Epic, eClinicalWorks, athenahealth and Allscripts. The product copy describes one click documentation back into the chart, and data ingestion spans the EHR, health information exchanges, claims and care gap files.

The depth likely differs by system, and the listing is vendor written.

The privacy policy states where data rests: AWS, North Virginia, United States, and the policy covers the personal data of customers' end users, patients included. The product runs as a hosted service surfacing insights inside the customer's EHR rather than as a separate installation.

What is left open is isolation and the processing path. Nothing says whether customers are separated in their own tenants, which services beyond the AWS host process patient data (including whatever generates the drafted documentation), or where the staff of Navina Technologies Ltd, the entity named in the privacy policy, access data from. No uptime commitment or implementation timeline is published.

No price, pricing unit or tier structure is published. How the product is sold is discoverable: four modules (Clinician Copilot, Risk Adjustment, Quality Management, Analytics) sold to provider groups, ACOs and MSOs, and health plans. Nothing says whether it is licensed per clinician, per attributed patient, per practice or as a share of the risk adjustment gains it produces. The last of those would change the vendor's incentives in a way a flat license does not, and performance linked pricing is common in value based care tooling.

The benefit side is better evidenced: an independent study by the American Academy of Family Physicians reported a 30 percent reduction in chart review burden and a 23 percent reduction in burnout, and the product was named 2025 Best in KLAS for clinician digital workflow. There is third party evidence of benefit and nothing about cost.

It is not stated whether any part of the fee is contingent on measured improvement in risk adjustment scores.

Focused on value based primary care and the organizations bearing risk in it: provider groups, ACOs and MSOs, and health plans. The value based tie is the product itself: HCC risk adjustment and RAF accuracy, HEDIS quality measures and care gap closure, and analytics tracking value based performance, with the FAQ covering risk adjustment audits for Medicare Advantage plans. The company has extended into ambient transcription combined with historical data, widening from a pre visit tool toward the full encounter.

Coverage is named with outside evidence behind part of it: the American Academy of Family Physicians study was run in primary care. Hospital, inpatient and specialty referral settings are outside the core positioning.