Matic

The platform is the model layer rather than a product with AI added to it. Ambient capture, note generation, coding recommendations, summarization across fragmented records, decision support and workflow routing are all model output running on what the company describes as a multi agent architecture. It is not an EHR with an AI feature; it is intelligence sold to sit inside somebody else's EHR or alongside one. No human scribe tier or services layer is described.

A review gate is stated as well as implied. Notes are described as ready to sign and send, coding is presented as recommending the right CPT, ICD-10 and evaluation and management codes rather than assigning them, and the terms of use require the physician to review the note and make any changes before submitting it to the medical record.

Nothing is published about what happens at the edges: no confidence threshold, no abstention behavior, no escalation path, and no statement of which outputs require sign off versus which flow onward automatically. That matters here because the platform reaches well past documentation into revenue cycle, inbox management and follow up, so the number of places where model output can move without a person looking at it is larger than for a pure scribe.

The architecture labels are present but thin, and the numbers actively undermine confidence. Multi agent system, multimodal input and adaptive templates are feature bullets rather than technical disclosure, and no model card, named model, evaluation methodology or error analysis exists.

The accuracy figure is given as 99 percent on the home page and has appeared as 98 percent elsewhere, which cannot both be right, and neither carries a denominator, comparator or definition of what counts as an error. The claim of 100 percent same day chart closure is an absolute rather than a measurement, since a rate of exactly 100 percent across any real population admits no exceptions. Support for more than 90 languages is asserted with no per language performance disclosure.

Hosting parties are named and the model layer is not. The privacy policy lists Google and Amazon as data cloud servers, SOUPEO as a further data cloud server, Google again for analytics, and Stripe, LiveChat and Mailchimp for payments and communications. No model, model family or foundation model provider is named, so where data is stored is known but not which model sees the encounter.

One deployment route answers the question structurally. Where the product is sold as embedded intelligence inside a record platform, the company states that data and control stay in platform, which for that channel means patient content does not accumulate in a separate vendor estate. The direct product uses the hosts above.

The privacy policy also says customer data is used to improve existing services and develop new products, with no statement on whether that includes training models or which data is excluded.

External assessment exists, though it stops short of clinical outcome data. KLAS Research has published a report specifically on Scribematic, which is assessment by a competent independent party with a real methodology.

Precision matters here: this is a KLAS report, not a Best in KLAS award. A named customer also exists, Catalyst Health Group, with its Dr Christopher Crow appearing in a published spotlight, which is more than a testimonial.

No peer reviewed study, controlled evaluation or outcome dataset is published, the KLAS report sits behind KLAS's own access gate, and the logo wall carrying Mayo Clinic, Epic, athenahealth and Privia does not state which of those are customers and which are integrations or platform relationships.

A retention schedule is published, with a development use permission instead of a training commitment. The privacy policy states that all transcribed data, including the health information in it, is deleted automatically after 30 days, held a further 30 days for accidental deletion recovery, and then permanently removed; users must copy notes into their own record system within that window. It separately says recordings are retained for 60 days, which does not match the 30 day rule for transcribed data, and it does not say which governs audio. Processing is encrypted at rest and in transit, and access is restricted to need to know under written permission.

The training half is not a safeguard. The policy says customer data is valuable in improving existing services and developing new products, and that data no longer needed is deleted or anonymized, without saying whether recordings or transcripts train models or how anonymization is done. No safety engineering is published: no guardrails, hallucination mitigation or handling of a safety event.

On the embedded route the company states that data and control stay in the EHR platform, which changes the retention picture for that channel.

Compliance is claimed and a business associate agreement is linked. The company's legal and privacy page, carried over from its earlier Scribematic brand, states that it complies with applicable HIPAA regulations and links both a HIPAA statement and a business associate agreement from its privacy policy, alongside compliance claims for CCPA, GDPR and PIPEDA.

The agreement's terms, tier and subprocessor coverage are not set out on the page, and nothing says where the agreement chain runs for the embedded EHR route, where the contracting party may be the EHR vendor rather than Matic.

Recognized certifications are named in the vendor's own material, with the question of whose they are left open. The site displays SOC 2 and ISO 27001:2022 badges and describes continuous monitoring, regular third party security audits, AES-256 encryption in transit and at rest, multi factor authentication and role based access, and it links a trust center at a dedicated subdomain.

Two things remain open. The privacy policy's own certification list, which includes SSAE 16 and ISAE 3402 Type II, ISO 27001, ISO 27017, ISO 27018, FedRAMP and PCI DSS, is attributed to the company's security certified data cloud server, which describes the host's certifications rather than the company's. And the site calls SOC 2 a certification without giving the report type or period, when SOC 2 is an attestation and Type I and Type II are very different assurances. The trust center may make the reports available without a request.

No clearance claimed and none required for the documentation component. No United States device pathway attaches to drafting a note that a licensed clinician reviews and signs, and the signature is the control point. It is not a simple absence, because documentation is one of four agents, and the other three each cross a different boundary.

Codematic is described by the vendor as delivering clinical decision support for accurate, compliant coding and payer reimbursement, and as aligning documentation to coding to protect revenue integrity. The vendor's own choice of words matters, because a company describing its coding product as clinical decision support has named the boundary rather than obscuring it. Proofmatic is described as bringing evidence based reasoning into clinical decisions in real time, which is decision support rather than documentation. Summatic answers questions over the assembled chart and returns cited answers pulled from source documentation, which is retrieval over the clinician's own record and a more traceable pattern than opaque inference, though it is still not documentation.

The platform thesis is care to claim: documentation informs coding, coding drives reimbursement, and the agents pass context between them automatically, a closed loop from clinical encounter to claim stated as the product design. Where documentation and revenue are produced by connected agents, nothing published says which artifact the clinician is actually attesting to when they sign.

The vendor also claims alignment with GDPR and PIPEDA. If the product is sold in the United Kingdom or the European Union, ambient documentation is treated in those markets as software as a medical device and a different regulatory conversation applies.

Mixed, with the coding language doing the damage. Intelligent coding is summarized as capturing revenue at the source, and the platform sells revenue cycle intelligence built on making every visit translate into clean claims, which is capture intensity framing. The platform position makes it structural rather than rhetorical: documentation, coding and claim all sit inside one system with no vendor boundary between them.

Two real counterweights belong alongside it. The phrase used for the outcome is fair reimbursement rather than maximized or optimized, which is more measured language. And codes are framed as recommended from what actually happened during the visit, which is the defensible framing.

Against all of it, the company publishes no fairness statement, subgroup analysis, accent or dialect performance breakdown or model limitations document, and support for more than 90 languages makes that absence consequential rather than academic.

Review before commit is stated in the terms of use: the company cannot guarantee accuracy or error free operation, users are responsible for verifying every transcription, and just as with a human scribe the physician must review the note and make any changes before submitting it to the medical record. That is a real mechanism for catching an error, and nothing stands behind the output beyond it. The privacy policy also asks users to hold the company harmless for damages from a third party breach.

The published numbers undermine each other. Accuracy is given as 99 percent on the home page and has appeared as 98 percent elsewhere, with no denominator, comparator or definition of an error. The claim of 100 percent same day chart closure is not a measurement, since any real population contains canceled visits, outages and encounters that run past midnight. More than ninety languages are supported with no per language performance, which is the figure that decides whether the product is safe to use with a given population.

Two claims sit side by side and only one of them is strong.

The weaker is EHR agnostic, works across any system, a universality phrasing that very likely means transfer rather than integration.

The stronger is a described and distinct product line. Matic sells embedded intelligence to EHR platforms, delivered through the EHR rather than around it, explicitly embedded not bolted on, running inside the platform's own workflows with data and control retained in platform and no rebuild required. That is a real architecture and a deep form of integration, because the scribe becomes part of the record system rather than a visitor to it.

The site does not say which of the displayed logos, athenahealth, Epic, Mayo Clinic, Privia, UrgentIQ, ChartPath, Lacenta and CHG, represent shipped embedded integrations as opposed to customers or prospects, and no API, standard or connector is named anywhere.

Broad platform coverage plus one differentiated option. The product runs on Windows, macOS, iOS, Android, web, desktop, mobile and tablet, so a clinician moving between workstation, ward and phone stays in one system. The differentiated option is the embedded route, where the company states data and control stay in platform, meaning for that deployment patient content remains inside the EHR vendor's environment rather than moving into Matic's.

For direct deployments the privacy policy names the hosts, Google, Amazon and SOUPEO as data cloud servers, and states a 30 day automatic deletion of transcribed data. No hosting region, residency guarantee or on premise path is published, and compliance claims spanning GDPR, PIPEDA and CCPA imply multi region operation without stating where anything actually sits.

No price, unit of pricing, tier structure or trial is published for any of the three buying routes, and every path leads to a booked demo.

A platform sold simultaneously to physician groups, to independent clinicians and to EHR vendors as embedded technology will carry three quite different commercial structures, and none of them is described. A per clinician rate and an embedded licensing fee are not comparable numbers.

Broad and adaptive rather than deep in any one place. Documentation is described as specialty tuned with templates that apply specialty specific formats automatically and adapt by both specialty and individual provider, with consistent output across sites, which is the requirement for a multi site physician group.

Settings named span clinic, hospital and on the go, and the customer and partner signals point at primary care through Catalyst Health Group, urgent care through UrgentIQ and post acute through ChartPath. More than 90 languages are supported. No specialty count is given and no clinical instrument, form or scoring system is named anywhere, and naming the instrument is what separates domain work from a specialty tuned template library.