DigitalOwl

The model output is what is sold. The products are AI-generated medical summaries (View), flags across bulk case files (Triage), a chat that answers questions over the record (Chat), automated decision trees over the full medical dataset (Workflows), case level inferences such as whether a condition progressed (Case Notes) and structured data delivered by API (Connect).

Remove the models and the platform has document intake and a portal left, with nothing a carrier or law firm would buy it for. Record retrieval, the part that does not depend on the models, comes from ChartSwap, the separate retrieval business it is being combined with.

The home page sells summaries that are complete: "quick and accurate medical summaries tailored to your line of business that ensure no detail is missed". No review mechanism can back that promise, because an omission cannot be seen in a summary and a buyer who believes nothing is missed will not reread the file.

The same page offers to "fully automate" triage by flagging predefined data points across a bulk of cases, and to structure records for machine to machine processing that fully automates the process, so outputs can feed decisions with no reader in between. What the product does offer is click-to-evidence, linking each finding to its source page, which lets a reviewer check what was said but not what was left out.

The company's terms say the opposite of the home page: it makes no warranty that results will be accurate or error-free, and the customer is "fully responsible to review and validate the service results".

The products are described by function and nothing that produces the output is identified. The terms say the service uses artificial intelligence and natural language processing, and the home page describes summaries, flags, chat answers and decision tree outputs, each linked back to the source page. No model, supplier, version, training data or update practice is named, and the 97 percent accuracy figure has no method. The only further material offered, a regulation white paper on the trust page, sits behind a download form.

Hosts are named and no model provider is. The trust page says the software and customer data reside on Google Cloud Platform and Amazon Web Services inside a dedicated virtual private cloud network, and that client data is not shared with partners or third parties.

The terms describe the service as using artificial intelligence and natural language processing but name no model, supplier or subprocessor, and no subprocessor list is published; the privacy policy's processor language covers the website. Whether any outside model provider sees medical records, and what the two cloud providers each do, is not established. The company was acquired by Datavant in 2025, according to reporting on the deal, and says it is becoming ChartSwap Insights, so the set of parties may change.

Vendor reported figures with no method, denominator, baseline or period. The home page claims precision of 97 percent or higher, review time reduced by up to 72 percent, and thousands of pages processed in an hour or less. Customers are described as leading insurance carriers and law firms; the site links a case studies library, but no customer is named on the home, trust, privacy or terms pages. Reporting on a 2023 funding round describes a strategic partnership with Reinsurance Group of America, which those pages do not mention.

The assurances are general and leave the central stewardship questions open. The trust page says client data stays within the services and is not shared with partners or third parties, is encrypted, and can be deleted or retrieved through the portal or API. The terms let the company compile anonymized, aggregated statistics from customer data and hold a perpetual license to the customer data within them.

Nothing states how long medical records and extracted data are kept, whether customer records are used to train or improve the models, or how de-identification is done for the aggregates. The people whose records are read are applicants, claimants and injured parties who did not choose the vendor and will usually never know it read their file.

HIPAA compliance is claimed and no privacy document reaching the service stands behind it. The home page says the company upholds strict compliance to HIPAA and SOC 2, and the trust page lists HIPAA among its certifications and attestations. The privacy policy (effective 1 January 2024) covers the website and is issued by an Israeli entity; the terms reach the service but say nothing about health information, business associate status or an agreement.

No business associate agreement is referenced in the published material. The role question turns on the buyer: the site names life and property and casualty carriers, claims organizations and law firms, most of which are not HIPAA covered entities and obtain records under an applicant or claimant authorization, so the vendor may hold no business associate role for them at all, while a health plan affiliated disability or long term care buyer may need one. The company does not say which role it holds for which buyer.

The trust page, linked from the footer as the Trust Center, names a SOC Type II covering the security, confidentiality, privacy and processing integrity criteria, presented as the company's own. The auditor, period and scope are not stated and no report or portal is offered, so currency and scope are the buyer's questions.

The page also describes encryption in transit and at rest, regular security audits, round the clock logging and monitoring, role based access and hosting in a dedicated network on Google Cloud and AWS. No HITRUST or ISO 27001 is claimed.

The output informs underwriting, claims, audit and litigation decisions, not a diagnosis or treatment, so no device claim is made and none applies. The regulatory exposure sits with the buyer's own regime, and the company says so only in general terms, offering a regulation white paper behind a download form and saying it tracks the evolving regulatory landscape.

The regimes in play are state insurance department oversight and unfair claims practices rules, state laws and NAIC guidance on insurer use of AI and predictive models, adverse action and disclosure duties when an AI informed decision affects an applicant, and evidence rules where output is used in litigation.

Transparency is committed to in general language with no evaluation behind it. The home page promises click-to-evidence and "no black box", and the trust page speaks of prioritizing transparency and meeting legal and ethical requirements; its regulation white paper sits behind a download form. No bias or fairness evaluation, subgroup result, error taxonomy or governance process is published openly. The exposure is concrete.

Outputs can feed automated triage and decision trees, and post-issue audits look for misrepresentation in already issued policies, where an inference that a condition predated an application can lead to a contested claim or a rescinded policy. Extraction quality also varies with the records themselves, handwriting, scan quality and the provider systems they came from, which the terms acknowledge for scans.

A mechanism lets a reviewer check an output, and the contract puts the result on the customer. Click-to-evidence links each finding to its source page. The terms warrant only that the services are provided with a commercially reasonable level of care and skill, state that some errors occur (attributed to poor scan quality), disclaim any warranty that results are accurate, complete or error-free, make the customer fully responsible to review and validate results, exclude liability for the customer's reliance on them, and cap liability at six months of fees; the only indemnity toward the customer covers intellectual property claims.

The home page's 97 percent accuracy figure has no method or denominator, so it does not amount to a published error rate. The person affected is an applicant, claimant or beneficiary with no relationship to the vendor and no route to correct what was extracted.

Integration is claimed with no system named. Connect is described as a data API that structures record data for customers' existing systems and lets them fetch data at each layer of the process, and the terms mention an API alongside the portal; the platform takes in documents of any kind, from record system exports to handwritten notes.

No underwriting, claims or case management system is named, no integration documentation is published, and there is no clinical record system connection, which its buyers would not expect. The company says it is becoming ChartSwap Insights, joining its analysis to ChartSwap's record retrieval network, which would give it a named source for records once delivered.

A single hosted service with location implied rather than committed. The trust page says the software and customer data reside on Google Cloud and AWS in a dedicated virtual private cloud, and the terms license the service for internal use in the United States, but no region, residency commitment or split between the two clouds is stated.

The privacy policy is issued by an entity headquartered in Israel and discusses EU transfers, which leaves open whether any staff or processing outside the United States can reach customer records. The move to ChartSwap Insights adds record retrieval, so the combined business will request and hold source records it did not hold before.

The main site carries no price, and the sales model is plain: an enterprise platform for insurers and law firms sold through a demo, and a self serve plan for legal teams reviewing under 30,000 pages a month, whose own page may state a price.

The unit of charge is not stated, whether per file, per page, per seat or per decision, and those behave very differently across underwriting, claims and mass torts; nor is it said whether record retrieval will be bundled or charged separately once the business becomes ChartSwap Insights.

Coverage is broad across the lines the company names for its own buyers: life insurers (underwriting, post issue audits, life claims, disability claims and long term care claims reviewed for material misrepresentation), property and casualty carriers (bodily injury, workers compensation and settlement demand package analysis) and law firms (personal injury, mass torts and medical malpractice), with an enterprise platform and a self serve legal plan for reviews under 30,000 pages a month.

None of these buyers is a care provider; the records are read to decide coverage, claims and litigation. Depth within each line is asserted rather than shown. The stakes differ by line, since underwriting and post issue audits act against the applicant while claims and legal work act on a claimant's injury.