Sample Healthcare

Language models drive the product's core work, inside a workflow platform that carries value of its own. The copilots use what the company calls proprietary clinical language models, adapted per customer from base models, to extract, classify, summarize and reason over medical records, correspondence and prior authorization documents, and that capability does not exist without them.

Around it sit workflow design that connects systems, teams and models, connectors to data warehouses, prior authorization intermediaries and a fax network, custom code, error monitoring and versioned lineage, much of which is orchestration rather than inference. Since April 2026 the platform is also sold inside TELCOR's laboratory revenue cycle system as TELCOR AI Systems, where it extends a billing product that stands without it.

Oversight is described as a phase, and nothing defines how it ends. The company says it starts by making a team ten times faster, then its systems learn from the team, and eventually they fully automate while the customer stays in control; the home page offers to monitor errors and move to full automation when the customer is ready. For prior authorization it says accuracy is maintained through expert review, and the acquirer's page describes human oversight at every step.

Those two descriptions do not match, and no threshold, error rate or sign-off defines when a workflow moves from reviewed to unattended, which workflows may, or what routes to a person after that. Audit trails and lineage let an organization reconstruct what the system did; they do not stop it.

The approach is described and no model is named. The home page says the copilots run on proprietary clinical language models, that customers can fine-tune models on their own data by adapting base models to their terminology and documentation patterns, and that performance is measured against clinical benchmarks before deployment. Versioned lineage graphs, anomaly alerts and audit trails that trace any value back to its source are published infrastructure.

What is missing is the model itself: no base model, supplier, version or update practice is named, no benchmark or result is published, and the 90 percent accuracy figure has no task, method or denominator.

Sample Healthcare explains how its models are built but names no model provider. The home page describes proprietary clinical language models adapted per customer from base models, without saying whose base models or where inference runs, and no hosting provider or subprocessor list is published. The parties the company does name are data connections: Kno2, Careviso, Glidian, Snowflake, Databricks and Salesforce. Since April 2026 TELCOR owns the company and runs the product inside its own platform, which adds a party to the data flow.

One named customer and vendor reported percentages with no method. A dated case study (May 2025) names Adaptive Biotechnologies on prior authorization, listing a 64 percent authorization improvement, 72 percent lower headcount and a two-week deployment. A second case lists a 62 percent authorization rate improvement and one fifth of expected staffing with no customer named, and a third card still carries placeholder text.

The home page adds 30 percent reduced costs, 90 percent accuracy and ten times faster teams, and the acquirer's page up to 75 percent faster document processing. None carries a baseline, volume, period or measurement method, so the figures cannot be tested.

Training use is stated for the customer's own models and nothing else is. The home page says customers customize models using their proprietary data, that custom models learn from their workflows, that models are tailored to each organization, and that the customer owns and controls every solution built together. It does not say whether customer data also improves shared base models, how long records, documents and outputs are kept, or what is de-identified.

No privacy policy is linked from the site, though its newsletter form refers to one; the acquirer's privacy policy is linked from its own pages. The material is clinical: medical records, intake documents and appeal files.

HIPAA compliance is claimed and nothing is published behind it. The technical foundation section of the home page lists the platform as HIPAA-compliant, and the observability section says audit trails satisfy HIPAA, SOC 2 and FDA record keeping. No business associate agreement, service privacy notice or statement of the role the company takes is published, and the site links no privacy policy at all.

Business associate status is structural for the work described, since the platform reads medical records and submits prior authorizations for providers and laboratories, and since April 2026 the question also runs through TELCOR, which owns the company.

The company presents a SOC 2 Type II attestation as its own. The technical foundation section of the home page lists SOC 2 Type II under certification beside a HIPAA-compliant claim, and a separate section describes audit trails, anomaly alerts and versioned lineage. No period, auditor, scope or report is offered, and there is no security page or trust center.

The same home page still carries template placeholder text in its outcomes section, and the about page in its team section, which leaves the attestation's currency and scope unclear, along with whether the report covers the platform as it now runs inside TELCOR.

No FDA position is stated, and most of the work is administrative: document extraction, intake, eligibility, prior authorization and appeals. The copilots also produce patient summaries described as improving care and route key information from medical records to the right providers, which is summarization for a clinician rather than a recommendation. Answering to the rules that govern the work is not a regulatory pathway.

Those rules are payer documentation and prior authorization requirements and, for laboratory customers, the billing regime TELCOR's platform serves; the home page's reference to FDA record keeping describes audit logs, not device status.

A stated evaluation step and nothing published from it. The home page says customers measure the performance of their tuned models against clinical benchmarks before deploying, and can monitor errors before moving a workflow to full automation. No benchmark, result, error rate by task, governance framework or subgroup analysis is published, and the 90 percent accuracy figure has no method.

The groups that matter for this product are document types and payers rather than demographics: handwritten or faxed records, scanned appeals and non-English correspondence are where extraction degrades, and prior authorization outcomes vary by payer.

Mechanisms exist to check an output, and nothing published stands behind one. Audit trails trace any value back to its source, versioned lineage graphs show which data and system version produced a result, and prior authorization work is described as keeping expert review. No terms of service, warranty, accuracy commitment or remediation is published: the site links no legal pages, and the 90 percent accuracy figure on the home page has no method or denominator. The consequence of an error falls beyond the customer: a prior authorization or appeal submitted with a wrong extraction delays a patient's test or treatment and becomes part of the payer's record.

Named connectors, none of them a clinical record system. The home page names connections to Kno2 for receiving faxes and communications, Careviso and Glidian for requesting and checking prior authorizations, Snowflake and Databricks for pulling order data, and Salesforce for syncing data, and says copilots connect to EHRs and insert results into the system of record. Since April 2026 the product also runs inside TELCOR's laboratory revenue cycle system. No record system is named, and nothing shows which connectors run in production at which customers.

Nothing is published about where the platform runs or where data rests. No hosting provider, region, tenancy model, residency commitment or customer hosted option appears on the home, about or outcomes pages or on the acquirer's product page. The statement that the customer's data, models and intelligence are always under its control, and the forward-deployed approach, describe ownership and an engineering engagement, not infrastructure.

Data also leaves through several connectors, including prior authorization intermediaries and a fax network, each holding records in its own right, and the platform now also runs inside TELCOR's system.

No price is published. The product is sold by demo, standalone or as part of TELCOR's platform. The acquirer's product page describes its model as a client-owned license with no percentage of collections, set against revenue-share competitors, which describes TELCOR's revenue cycle offering rather than a stated price for the AI workflows.

The unit of charge (per workflow, per document, per authorization or license) is not stated, nor whether the price changes as workflows move to full automation or when the product is bought through TELCOR.

Coverage is named, with validation behind part of it. The company names its customers as molecular diagnostics companies, oncology groups and laboratories, and the acquirer sells the product to independent laboratories and to hospitals and health systems. Workflows named are medical record abstraction, patient intake and eligibility, prior authorization, appeals, payer follow-up and correspondence.

Prior authorization at a diagnostics company is supported by a named, dated case (Adaptive Biotechnologies, 2025); the intake, abstraction and appeals claims rest on unnamed customers. Its owner, TELCOR, sells appeals and payer follow up to laboratories as revenue cycle work, which places the product in that market too.