Onpoint Healthcare Partners

The model drafts the work and people finish it, so the AI is the engine of the core module rather than the whole product. The company describes an 80 percent AI and 20 percent expert in the loop split: ChartFlow builds the note, the chart prep and the medication and problem list reconciliation from the encounter and the record, and clinical auditors review the result before it reaches the provider.

A 2023 interview with the then chief executive described the same shape, with the model doing about 70 percent of the note and a team finishing it. The HCC coding page credits certified human coders with the accuracy figure it publishes, and the RCM line is a service.

Remove the model and a documentation, coding and revenue cycle service business would remain, which is why this sits at B rather than A. The charting capability as sold today does not exist without it.

Ask what share of encounters reach the provider without auditor edits, and whether that share is tracked by specialty.

The oversight structure is described and its thresholds are not. AI agents draft and act, clinical auditors review roughly a fifth of the work, and the provider reviews and signs. SignalsFlow pends orders, visits and referrals for sign off inside the EHR rather than placing them. A 2024 interview added that the level of human involvement is matched to the complexity of each encounter.

What is missing is the routing rule. Nothing says what sends a draft to an auditor, what share of notes skips human review, or what happens after an error reaches the record.

The marketing pulls against the design. The site says the AI acts rather than assists and calls the platform fully autonomous, while the same pages credit the expert in the loop for its accuracy. A customer quote on the home page describes reviewing notes in the morning and accepting them because they are usually perfect, which is exactly the review behavior a routing rule has to anticipate.

Ask what triggers auditor review, whether every note is audited or a sample, and how errors found after signing are handled.

The architecture is described in general terms and nothing current is identified. The site calls the platform agentic and generative AI that employs all modalities, with AI agents and clinical auditors working together, and names no model, version, training data or update practice.

A 2023 Healthcare IT Today interview had the then chief executive saying ChatGPT was built into the IRIS Virtual Clinical Assistant, and a 2024 interview placed the system on Microsoft Azure. Together those suggest an OpenAI model served through Microsoft, but the vendor states neither on its current site and the naming predates the current product set, so a buyer should not assume it still describes ChartFlow.

Ask which models produce notes, codes and orders today, how model changes are communicated to customers, and whether outputs are pinned to a version.

The human half of the chain is disclosed and the model half is not. The site names the delivery entities where reviewers work: Onpoint Delivery Partners Pvt. Ltd. in Hyderabad, and Medico Escribatorios Popular De Mexicali, S. de R.L. de C.V., in Mexicali and Monterrey. For a buyer with residency or subcontracting obligations, that is the fact in this chain to settle first.

No model provider is named on the current site and no subprocessor list is published. A 2023 trade press interview had the then chief executive saying ChatGPT was built into the product, and a 2024 interview placed the system on Microsoft Azure, but neither is a current disclosure and neither states retention or training terms.

Ask for the subprocessor list, the model provider and the terms that govern encounter data once it arrives there, and which delivery entity handles your work.

The headline claim is 99.5 percent clinical accuracy, achieved by combining AI with clinical auditors, and the published evidence does not reach it. The figure appears on the home and ChartFlow pages with no method, denominator, reference standard or period. The same holds for the other figures: up to 36 charting minutes saved per patient, 3.5 hours a day per provider, 99 percent provider satisfaction, and a 90 percent cut in documentation time at an unnamed academic medical center's oncology service. A 2024 interview gave 98 percent for the same platform.

The strongest outside evidence is KLAS. The platform was named to the KLAS 2025 Emerging Solutions Top 20, placing third for clinician experience and fifth for outcomes, and in June 2026 KLAS published an ROI validation at Southwest Care, a federally qualified health center in Santa Fe and Albuquerque. Those are customer based assessments rather than an accuracy study, and the press release announcing the validation gives no figures.

The accuracy claim also measures the combined output of model and auditors, so it says nothing about the model alone.

Ask for the validation report from KLAS, the method behind the 99.5 percent figure, and the auditor edit rate.

Only general assurances are published, and they do not answer the questions this axis asks. No retention period for audio, transcripts or drafts is stated, nothing says whether customer encounters train the models, and nothing describes de identification.

The one statement bearing on training comes from a 2024 interview in which the former chief executive described a continuous learning loop that improves the platform over time using real world data. That suggests customer data feeds model improvement. It may be governed by contract, but nothing published says so.

The human review step adds an exposure the software alone would not carry: clinical auditors, some working from delivery centers in India and Mexico, read protected information in drafts.

Safety engineering, meaning guardrails, hallucination mitigation and how a safety event is handled, is not published, as it is not for any record in this index. The vendor's answer to hallucination is the auditor layer, which is an oversight control rather than safety engineering.

Ask for the retention schedule, the training position in writing, and how auditor access to PHI is limited and logged.

Compliance is claimed and nothing underneath it is published. The What We Do page lists HIPAA Compliant among its awards and certifications, and the former chief executive described HIPAA as a design priority in a 2024 interview. No business associate status statement, agreement or scope is published.

There is also no privacy policy on the estate. The website terms say a Privacy Policy is incorporated by reference, but no such page is linked in the footer or listed in the site map, and the terms themselves govern use of the website only.

For a service whose human review workforce is partly in India and Mexico, the scope of the agreement matters more than usual, because it decides whether the offshore delivery entities are bound as subcontractors.

Ask for the business associate agreement, confirmation that it binds Onpoint Delivery Partners in Hyderabad and the Mexican delivery entity, and the privacy policy the terms refer to.

A recognized certification is named and nothing behind it is retrievable. The What We Do page lists SOC 2 Certified among its awards and certifications, with no report type, period, auditor or scope, and there is no trust portal or security page on the estate.

A 2024 interview muddies the scope. The former chief executive said the company built on Microsoft Azure to ensure it was high trust and SOC 2 certified, which reads as describing the cloud provider's certifications rather than Onpoint's own. The site badge is graded as the vendor's own claim, with that scope question left for the buyer to settle, along with whether the audit covers the offshore delivery entities where reviewers work.

Ask for the SOC 2 report, its type and period, and confirmation that its boundary includes the review workforce in India and Mexico.

No device claim is made, and the product the record leads with is documentation and coding, which needs none. The site states no regulatory position.

One module reaches further. SignalsFlow describes AI driven signal surveillance that watches remote monitoring data and surfaces deterioration early, using what the vendor calls FDA compliant connected devices, with US based state certified nurses managing patients. The devices' clearances belong to their manufacturers and do not cover the surveillance layer. Nothing published states whether that layer is positioned as decision support, what it alerts on, or what basis the nurse is shown.

Because it is a side module sold for a different workflow, it is named here rather than graded, and the grade follows the charting product.

Ask which devices are used and under whose clearances, what the deterioration signal is based on, and whether a clinician reviews it before any action.

Governance is asserted through a partner, and no evaluation of model behavior is published. The What We Do page lists ALIGNMT AI Certified among its certifications. ALIGNMT AI is a New York governance platform that monitors AI systems against frameworks such as ISO 42001 and the NIST AI Risk Management Framework, and Onpoint's chief operating officer was quoted in its April 2025 launch describing real time monitoring of Onpoint's AI. The former chief executive also cited involvement with an organization focused on ethical AI practice.

None of this publishes what was tested, against which criteria, or with what result, and no subgroup performance is published.

That gap matters here. ChartFlow markets multilingual support and the company names English and Spanish, while nothing shows whether accuracy holds across languages, accents or the safety net populations its health center customers serve.

Ask for the ALIGNMT report or report card and what it covers, and for accuracy results broken out by language and patient population.

Review mechanisms exist, and nothing published stands behind the output. Clinical auditors check drafts, providers review and sign, and SignalsFlow pends actions for sign off, which are routes to catch an error before it lands. The only published legal document is a website terms of use, which disclaims warranties and liability for use of the site and says nothing about the service.

The HCC coding page comes closest to a commitment. It credits certified coders with 95 percent average accuracy and promises a quality assurance process with regular reporting to ensure the benchmark is achieved, but it states no method and no remedy if the benchmark is missed. The 99.5 percent clinical accuracy figure has no method or denominator, so it is not an error rate a buyer can hold the vendor to.

Ask whether the 95 percent coding benchmark and the 99.5 percent clinical figure can be written into the contract with a remedy, and who answers for a coding or documentation error found after signing.

Named systems are claimed and one listing is verifiable. The solutions page says Iris integrates directly with athenahealth, eClinicalWorks, Epic, NextGen and others, and Onpoint announced a listing on the Veradigm Marketplace in 2025. A 2025 announcement describes the platform as an orchestration layer running inside leading record systems and standalone where needed.

SignalsFlow pends orders, medications and referrals for sign off inside the EHR, and ChartFlow works from record data for chart prep and reconciliation, so the design reads and writes. No page documents the direction, method or depth for any named system, and Veradigm is the only marketplace listing the vendor has announced.

Ask for the integration method for your EHR, whether notes and codes write back as discrete data or as text, and for references running on the same system.

One hosted service is implied and its location is not committed. The former chief executive said in a 2024 interview that the system was built on Microsoft Azure; no region, tenancy model or residency commitment is published.

The more consequential residency fact is on the site itself. The human review workforce operates from delivery entities in Hyderabad, India, and in Mexicali and Monterrey, Mexico, so protected information in drafts is handled across borders by people even if the software stays in the United States. Nothing states whether a US only review option can be contracted.

Ask for the hosting region, whether customer data is isolated, which steps of the workflow are performed outside the United States, and whether US only review is available.

No price is published, but the sales model is discoverable. The solutions page says the managed service is sold on subscription, with discounts applied by practice size, specialty and workload, and positions it against the cost of hiring staff.

Modules can be tried before buying. The ChartFlow and CareFlow pages offer a risk free trial, a provider ROI assessment and a demo, and an advertised offer adds a free onsite assessment.

A buyer can establish that cost scales with practice size and workload before making contact, but cannot size it. The human review layer and the offshore delivery that sit inside the price are not broken out. The solutions page also calls the pricing transparent, which describes the sales conversation rather than anything on the site.

Ask for the rate at your size, what share of the fee pays for human review, and whether modules are priced separately.

Coverage is named and partly supported. The vendor reports more than 2,000 providers across more than 35 specialties and names its settings, from federally qualified health centers to academic medical centers and independent practices.

Two deployments carry outside or published support. Southwest Care, a federally qualified health center in New Mexico, is the site of an ROI validation KLAS published in June 2026, and an unnamed academic medical center's oncology service is the subject of a vendor case study. Customer quotes on the site come from urology, acute care and multispecialty groups. ChartFlow lists multilingual support, and the company's KLAS profile names English and Spanish.

The record also appears in three other lanes because the vendor sells a dedicated module for each: CodeFlow for automated E&M, CPT, HCPCS and ICD coding; CareFlow and NetworkFlow for medication prior authorizations, referrals and care coordination; and CareFlow with ChartFlow for HCC risk adjustment, care gap closure and quality measure programs.

Ask which specialties have their own templates and audit staff, and which languages are covered by human review as well as by the model.