Clicks Health

Clicks runs administrative agents inside a healthcare team's existing software. The agent reads clinical records and payer correspondence, interprets what a request needs, fills forms and prepares supporting packages. Computer use lets it navigate screens through its own authorized account, with APIs used where available. For New Jersey injury claims it infers selected treatment form fields and drafts denial appeal rationales from the existing record.

These functions go beyond copying fixed fields between applications. Customers define the workflow and permissions, while the agent performs the reading, interpretation and execution that make up the service sold to revenue cycle teams.

The APTP agent waits for a locked encounter note and returns cases with missing fields or disagreements between orders and notes. If the note is not locked in time, the item goes back to the coordinator. The provider reviews and signs the prepared form; the agent does not decide medical necessity. Appeal packages likewise return to staff for review and submission. Customers choose the agent's account permissions, workflow and escalation rules.

Diagnosis codes come from existing clinician documentation, while six other form elements may be inferred. These controls distinguish reading a chart, preparing a draft and authorizing a final action.

Computer use gives the agent access to the same screens, keyboard actions and applications used by staff. A separate authorized account determines what it can reach, while APIs offer another route into supported systems. The agent reads scanned letters and comments, interprets payer requirements and assembles forms or appeal text. Some APTP values are inferred; diagnosis codes and most other fields are copied from existing records, with fixed rules governing dates and duration. No model provider, base model or version policy is identified. The disclosed method explains the execution layer more fully than the language and reasoning models underneath.

AI model providers are acknowledged as outside processors, subject to a contractual prohibition on training with customer data. Other supplier categories cover hosting, analytics, payment processing and support. The service uses hosted agents that interact with customer accounts in EHRs and payer portals, and processing is assigned to US or EU regions according to the customer. No supplier is named at the model layer, and the public trust center does not enumerate a subprocessor list.

Provider identity, downstream retention and the handling of model requests across borders therefore remain unspecified. Customer account permissions control access to source systems separately from those supplier terms.

An orthopedic group in New Jersey reports 50% more appeals won over two months and 160 appeals. Staff work fell from 25 to 40 minutes per appeal to about five, with labor cost dropping from roughly $20 to $5. The result combines more appeals attempted with changes to package preparation, rather than isolating the effect of draft quality. A separate APTP workflow reports 24% fewer administrative denials and removal of work previously occupying three staff members. Princeton Brain, Spine & Orthopedics is named as a customer, but the outcome stories do not identify their group or calendar dates. These remain vendor case results.

Customer inputs, outputs and uploaded documents are excluded from model training, fine tuning and improvement under a contractual commitment. The privacy terms also prohibit outside model suppliers from training on customer data. Processing follows the customer's region, and access is controlled through permissions and recorded actions. At contract end, data remains available for export for a reasonable period before deletion under the retention policy.

A fixed retention timetable, deidentification rules and an AI safety incident procedure are not specified. Operational safeguards include drafts, source records and exception handoffs, particularly when an authorization package contains missing or conflicting information.

HIPAA compliance is stated for the healthcare agents and is repeated in the service terms and public trust center. The service terms cover software used inside customer systems, with encryption, action trails and isolated processing environments. The privacy policy describes outside AI processing and limits provider training use, although its introduction is framed around website visitors.

Neither a business associate agreement nor a clear statement of business associate status is accessible. The HIPAA claim therefore does not establish the contractual scope for an organization's clinical records, insurer correspondence or documents passed through an agent's account.

Clicks names SOC 2 Type II as its own security assurance and describes annual audits and independent penetration tests. The public Vanta trust center shows controls for database authentication, encryption key access, asset disposal, vulnerability monitoring, disaster recovery and retention procedures. Its resource list contains an engagement letter rather than the SOC report itself. The security page adds encryption in transit and at rest, SAML sign on and IP allowlisting. Audit period, auditor and report scope remain unavailable. The 99.9% uptime service level concerns operational availability and does not establish the correctness of an agent's output.

The healthcare offer is administrative. Agents prepare authorization and appeal paperwork, retrieve existing diagnosis codes, record payer responses and move work between clinical and billing systems. Medical necessity and care decisions remain with qualified clinicians. For APTP forms, the provider retains attestation and signature, and the system does not add a new diagnosis code. Payer calls are part of the workflow, while direct patient contact stays with the customer's team. There is no device clearance claim for these agents. Their role is to execute insurance and office processes around an existing clinical decision.

Customer defined permissions, review points and escalation rules govern agent actions. An action trail preserves the evidence used and the decisions taken across record systems and payer portals. In the APTP workflow, missing information and conflicting orders produce a handoff, while the signed clinical decision remains with the provider. These controls establish accountability for work in progress.

Model testing across patient groups, specialties or payer populations is not described, and the security trust center's control checks concern information protection rather than evaluation of model behavior. The outcome examples measure administrative effort and appeals won, without subgroup comparisons.

A person can review an appeal draft against the supporting record before submitting it, and action logs preserve what the agent did. The service terms cap aggregate liability at fees paid during the preceding twelve months and exclude indirect damages. Customer indemnification applies to misuse or breach of the terms. A prior authorization guide offers a guarantee of the agreed outcome, but gives no public measurement, payment remedy or correction obligation.

That promise is separate from the 99.9% uptime target. The practical correction path is therefore staff review and exception handling, without an explicit remedy tied to an incorrect generated form or appeal.

Supported systems named by Clicks include Epic, Oracle Health, athenahealth, eClinicalWorks, ModMed, NextGen and MEDITECH. Agents can use APIs or operate through an authorized user account in the application itself. The authorization workflow reads records and orders, works through payer portals, then returns submission status and confirmation details to the EHR. Appeal packages are uploaded and assigned to the case owner.

The eClinicalWorks demonstration explicitly uses fictional patients in a test environment. No marketplace listing or named production implementation links a specific record system to those bidirectional workflows. Screen access is part of the offer, not necessarily a native connector.

Customer data is hosted and processed in the customer's region, with US processing for American customers and EU processing for European customers. The privacy terms also allow cross border transfers with contractual safeguards. Service terms describe isolated processing environments, but do not specify how one customer's tenant is separated from another or identify the cloud host.

Access combines a dedicated agent account in customer software with SAML sign on, role permissions and IP restrictions. Automated backups and redundancy support the hosted service. An onsite installation option is not part of the described delivery model.

Engagement begins with a selected workflow, process mapping and an agreed outcome. Clicks then builds a custom agent around the customer's existing tools, with a stated implementation target of five weeks or less. Pilots use actual processes and data before the customer commits. This is a configured automation engagement rather than a self serve subscription with a standard feature bundle. The current healthcare offer supplies no price, unit of charge or fee schedule. The $20 to $5 comparison in the appeal story measures staff labor per appeal; it is not the amount Clicks charges for its agent.

RCM companies, orthopedic practices, rural hospitals and other healthcare operators are the intended customers. The most specific operating examples concern New Jersey Personal Injury Protection requests for imaging, durable medical equipment and injections. Agents prepare APTP forms and appeals using accident records, treatment orders and progress notes.

Princeton Brain, Spine & Orthopedics is a named customer, supporting the orthopedic setting, while coverage extends commercially to referrals, eligibility, payment posting, refunds and disability forms. Outcome results for rural hospitals and other specialties are not broken out. Patients remain in contact with staff; the agent's calling workflow is directed to payers.