Clarus

Conversational agents now support inbound scheduling and outbound outreach, while transcription and summaries help staff process patient messages. These features extend an established platform that also offers rules based IVR, secure messaging, call routing, and on-call scheduling. A practice can choose a conventional routing experience or a conversational interface, so the software's operational value is not wholly dependent on generative AI.

The company timeline places the conversational launch in 2026, after years of call management deployments. The distinction matters when connecting older customer experience to the current agents: shared platform infrastructure does not make their behavior identical.

Urgent calls can reach an on-call clinician, while routine requests enter the next business day's work queue. Staff control schedules and fallback routing, and outbound concerns can transfer live or arrive as a flagged message with context. These controls preserve a human path for difficult conversations. At the same time, the current service promises resolution of every call and asserts that false alarms never wake providers.

Those unqualified claims leave the possibility of incorrect urgency judgments and missed escalations inadequately bounded. Configurable routing and clinician availability provide oversight, but neither establishes error free classification by the conversational agents.

The current products combine conversational AI, transcription, interaction summaries, and configurable routing. Inbound workflows can use rules based IVR or an agent that checks appointment availability and records the result. Outbound workflows use patient information and clinical criteria to select outreach and preserve conversation context for staff.

This explains how several software capabilities fit together, but the underlying foundation models, production versions, training data, and update review practices remain unspecified. A description of the workflow is therefore available without a technical account of how the language and speech models are selected, evaluated, or changed.

Patient calls pass through speech recognition, transcription, conversational responses, and storage, yet the product materials do not identify the model providers responsible for those steps. The website privacy notice permits third-party service providers to perform assigned tasks and limits unrelated use, but its scope is website information. A current service processor register, model provider terms, inference retention periods, and training restrictions remain unspecified. The available material therefore does not establish who can access clinical conversations throughout the processing chain.

The Graham Healthcare Group case describes more than 21,000 monthly inbound calls handled by eight receptionists before implementation. Smart routing and transcription were associated with a 33% reduction in inbound volume and improved follow-up visibility. A separate June 2024 announcement reports savings and shorter message resolution at Graham Residential Home Health. These are concrete customer accounts of established call management.

They do not supply a dated evaluation of the conversational agents introduced in 2026, including their accuracy, escalation errors, or current comparative outcomes. The newer product's performance claims therefore remain distinct from the older deployment results.

The inbound product specifies seven years of storage for patient interactions, including recordings and communication records linked with the EHR. This provides a definite retention period for records used in follow-up and audit review. It does not explain whether that period is a minimum or a deletion deadline, nor how copies held by speech or model providers are handled. Training use, deidentification practices, and deletion behavior after retention remain unspecified.

Recordings allow staff to inspect conversations, but the current information does not describe technical safeguards against fabricated responses or a dedicated process for investigating unsafe agent behavior.

Patient communication features carry a HIPAA compliance claim, with encryption, secure messaging, and retained interaction records. These controls are relevant because the platform handles identifiable patient requests and sends information into clinical workflows. The accessible privacy notice explicitly addresses website visitors and describes ordinary contact information, browsing logs, and third-party website services.

It does not establish the contractual treatment of clinical communications. An accessible customer BAA, service specific data rights, product coverage, and downstream business associate obligations remain unspecified. The website policy therefore cannot resolve the terms governing PHI handled by the inbound and outbound agents.

A SOC 2 audit claim accompanies the platform's HIPAA language and encryption commitments. It provides a recognized security assurance claim for a service handling patient calls, transcripts, and messages. The audit type, assessment period, covered systems, and access conditions for a current report remain unspecified. Encryption and retained communication trails describe operational protections, but they do not establish an independently verified scope for each conversational feature. The available information therefore supports the existence of an attestation claim while leaving its current coverage of the application and downstream processing unresolved.

The communication workflows include sorting urgent from routine concerns and initiating follow-up around discharge, procedures, and chronic care thresholds. Outbound agents capture patient responses and escalate issues that need clinical attention. These activities can influence the timing of care, even when clinicians ultimately decide treatment. The current product's regulatory position and the limits of its urgency assessments remain unspecified.

Routine scheduling and reminders are administrative tasks, but that narrower scope does not explain the clinical sorting functions. An on-call handoff identifies a responsible recipient without establishing how the underlying assessment is classified or validated.

The platform supports specialty specific workflows and configurable call routing, but those operating settings do not establish a formal process for governing model behavior. For the current conversational agents, named oversight responsibilities, release review criteria, subgroup performance results, and independent behavioral assessments remain unspecified. The available customer accounts primarily describe operational experience. They provide no comparable analysis of whether language, age, or patient population affects urgency classification or response quality.

The inbound product claims resolution of all patient requests in less than two hours, creating a measurable service assertion. Call recordings, transcripts, and dashboards also give staff a way to inspect communications and follow up on unresolved concerns. That visibility can support operational correction, but the product statement is not an accessible contractual remedy.

Service credits, reimbursement, indemnity for incorrect responses, and a defined process for contesting an agent's harmful output remain unspecified. The available recourse is therefore centered on staff review and a falsifiable resolution commitment rather than an established allocation of losses from AI errors.

The June 2024 Homecare Homebase announcement describes a bidirectional design: patient demographics would appear in Clarus, and messages would return to the medical record as coordination notes. Its language concerns planned integration delivery. Current product material names additional EHR platforms and describes automatic communication records and patient data synchronization. Together these support a recognizable integration approach beyond generic interoperability language.

A named current customer using the complete read and write connection remains unspecified. The presence of a customer in the older announcement establishes Clarus use, but does not by itself confirm that the announced integration was operating there.

The service is hosted in the cloud and accessed from desktop or mobile devices, with no local hardware installation required. Implementation covers workflow configuration, EHR integration, and staff training, and the company describes a typical launch within 30 days. Those details explain the buyer's setup and access model. Hosting regions, tenant separation, disaster recovery locations, and the processing locations for audio and language inference remain unspecified. Cloud access therefore establishes how staff reach the application without identifying where patient communications are stored or whether any model processing or support access occurs across borders.

Practices and larger organizations purchase through a sales conversation and demonstration. Customer accounts describe lower monthly costs or annual savings relative to prior answering arrangements, but those figures reflect particular customers' comparisons rather than a current price schedule. The base subscription, billing unit, usage allowances, and implementation charges remain unspecified. Product claims about time saved and increased revenue describe expected value; they do not establish the amount a buyer would pay for inbound service, outbound campaigns, or both.

Named customers span orthopedics, pediatrics, obstetrics and gynecology, pulmonary medicine, gastroenterology, and home health. Examples include Kansas City Bone and Joint, Vernon Pediatrics, and Graham Healthcare Group. The service also targets health systems, multisite organizations, rural clinics, and federally qualified health centers. These references support breadth in patient communication across ambulatory and home care settings.

They vary in scope, with many discussing routing, recordings, or staff workload rather than current conversational agent performance. Comparable specialty specific validation of urgency sorting and autonomous responses remains unspecified, so deployment breadth alone does not establish equal accuracy across clinical populations.