Charta Health

Large language models do the reviewing. They read clinical documentation, extract its meaning and run analysis modules configured for each note type, and Charta contrasts this with rules engines that depend on rules their users write.

Reviewing every chart before billing, rather than a sample, is the product, and Charta says that was too costly to be realistic before AI. Without the models there is nothing left to sell.

Three modes are offered: coding every note autonomously after it closes, correcting coding directly in the EHR, or sending problem charts to a person, with citations to the supporting documentation on each suggestion. What makes a chart a problem chart, and so what reaches a human, is not described.

The product pages pair that autonomy with absolute claims. Charta reads all documentation 'to completely and accurately code every service', promises to ensure 'you code every service delivered', and asks 'what if you could guarantee revenue integrity and clinical compliance on every chart?'. Promises of completeness and accuracy on every chart work against the human review that catches a wrong or missed code.

Per client tuning shapes each deployment. A Charta AI engineer configures analysis modules for each type of encounter note during onboarding, around the client's own review criteria, and the models read handwritten notes, faxes and images as well as structured EHR data, which older rules engines built on structured fields could not.

No model provider, version or evaluation method is given, and because each deployment is tuned to its client, performance at one organization says little about another.

Large language models tuned per client are the approach, and no party in the chain is identified: no foundation model provider, host or subprocessor list.

The GDPR compliance statement implies a record of processing activities exists, but no list of processors appears on the site, so the companies that receive complete encounter documentation are unknown to the client.

An 11 percent revenue uplift across pre billing chart review is Charta's headline figure, and named executives supply the rest: KidsCare Home Health reports a 25 percent cut in clinical management workflow duties, and Third Eye Health and Family Care Center report 90 percent more charts reviewed before billing. An FQHC case study found missed revenue on 25 percent of notes.

Pre bill coverage measures scope rather than accuracy, and none of the figures carries a baseline or outside audit. The one accuracy figure, 98 percent on payer risk adjustment coding projects, comes without a method, and none is given for provider coding, the work most of the revenue figures depend on.

The whole chart passes through the platform. Charta reads all provider documentation for every encounter, faxes and scanned images included, and tunes each implementation on the client's review feedback; its FAQ states HIPAA, SOC 2 and GDPR compliance.

Retention, de identification, whether client documentation trains models used for other clients, and which model provider receives the notes are not stated. Citations on each suggestion are the only safeguard against invented content that Charta describes.

HIPAA compliance is answered yes in the FAQ. No business associate statement, agreement or privacy document covering the service appears on the site, so the terms under which Charta handles PHI cannot be read before contracting.

Those terms would govern complete encounter documentation, since Charta reviews every note rather than a sample.

Security assurance comes by request. The FAQ states SOC 2 compliance and directs buyers to compliance@chartahealth.com for reports, without giving the report type, period or auditor.

There is no trust center to browse, so the scope of the SOC 2 work, and whether it reaches the model layer that reads client charts, is visible only once the report arrives.

Coding and chart review are administrative functions, and no device claim is made.

The exposure is program integrity. Charta verifies that E/M codes are billed 'at the highest defensible level' and can code without review, and when a government payer is billed, liability for an unsupported code falls on the billing provider under the False Claims Act. On the payer side it promises to make a risk adjustment program 'audit proof', a marketing line with no compliance position behind it.

How Charta monitors or tests model behavior is not described, and no governance framework or evaluation is offered.

The direction of error matters for this product. Charta measures its value as revenue uplift, 11 percent in RVUs per encounter, and verifies that E/M codes are billed at the highest defensible level, so a model tuned for capture can drift toward higher coding without any single chart showing it.

Each suggestion cites the documentation behind it, so an unsupported or invented code can be checked at review rather than found in an audit months later, and problem charts can go to a person.

No warranty, indemnity or correction commitment stands behind the output. Where Charta corrects coding in the EHR without review, the citation trail is the only check.

Integration runs both ways in Charta's account: the platform reads all documentation after a note closes and can correct coding directly in the EHR, and the FAQ says it has integrated with dozens of commercial EHRs and homegrown EMRs.

No EHR, standard or marketplace listing is named, so which systems support correction in place, rather than review alone, cannot be confirmed.

Charta runs as a hosted platform that reads EHR documentation after each note closes.

Host, region and tenancy are not stated. The GDPR compliance statement implies some processing falls under EU rules without saying where, and the language model layer adds a processing location the client cannot see.

Pricing goes through a demo request. Charta sells to provider organizations, health systems, MSOs and payers, builds each integration with dedicated engineering support, and measures its value as revenue uplift in RVUs per encounter.

Whether the fee is per encounter, per provider or a share of uplift is not stated, and a fee that scales with uplift would reward higher coding.

Fee for service providers across specialties make up the provider market: primary care, FQHCs, urgent care, behavioral health, provider groups, MSOs and health systems, with payers served for risk adjustment and HEDIS. Named customers cover home health (KidsCare Home Health) and provider groups (Third Eye Health, Family Care Center), and an FQHC case study covers community health.

Inpatient facility coding is not the target, although Charta says any organization with a reason to audit charts can use it.