Artera

The virtual agents, Flows Agents, and AI Co-Pilots are model driven and are what the company is now selling against. Underneath sits an eleven year old patient communications and messaging platform; AI is the current layer on a messaging infrastructure whose value stood before it. The vendor's own framing is instructive: its CEO argues that building AI agents is becoming commoditized and that the moat is domain experience, data, and distribution rather than the models.

The hand-off structure is described, and the rule that triggers it is not. Artera's scheduling agent answers calls around the clock. Agents route complex requests to staff with full context, a warm handoff to a person is demonstrated, staff can take over any conversation and see its full history, and transfer lines are configured for each customer.

The criteria that send a caller to a person, by channel and hour, are not published, although one customer story describes agents recognizing alarm symptoms and escalating to on call clinical staff. The agents are also sold as remembering every interaction, preference and conversation, which sets the scope of what they carry between calls rather than promising a correct result. The platform material says agents identify issues and route around problems without manual intervention, and that the system optimizes its own scheduling workflows.

Capabilities are described, and the technology beneath them is not named. According to its own pages and marketplace copy, Artera's platform coordinates agents across voice, text and web for scheduling, intake, referrals, billing and care gaps. It carries patient memory across conversations, AI and LLM-driven translation in 109-plus languages, real-time sentiment tracking and staff co-pilots for translation and summaries.

The solutions page says its AI monitors, improves, tests and deploys itself, flagging upgrades for human review when appropriate. No model, architecture, version or change notice is published, and nothing says whether the language models are Artera's own or licensed. Audio demonstrations are labeled as mixing simulated interactions with real de-identified calls.

Artera hosts on its own infrastructure and names no model provider. Its trust center says customer data sits on Artera's servers in onshore data centers with SOC 2 attestations, and its marketplace copy describes LLM-driven translation, but no model provider, foundation model or subprocessor list is published.

Two kinds of party necessarily sit between a patient record and a message: the telecommunications carriers and messaging aggregators that carry every text and call, and whichever language model provider performs translation and conversation, if not Artera itself. The stated exclusion of identifiable patient information from training says nothing about what a third-party model provider retains.

One named, dated deployment with figures, and several unnamed ones. Atlantic Health used Artera's agents to call patients a week before scheduled colonoscopies; over 30 days from February 2026 the company reports a 38 percent reduction in staff time on manual outreach, with 43 percent of patients who answered confirming their identity and 39 percent confirming their appointment.

Other home page stories are unnamed: a practice with 43,000 conversations automated, 1,250 staff hours reclaimed and a 67 percent drop in call abandonment within 60 days; an appointment confirmation agent credited with 15 percent better confirmation and 14 percent fewer no-shows; and a gastroenterology practice's symptom triage described without figures. MEDITECH publishes a hospital case on reduced no-shows using Artera through its Expanse Patient Connect product.

The company also cites 65 percent AI agent task completion, more than 2 billion interactions a year and a 2026 Best in KLAS ranking, which measures customer opinion rather than outcomes. No baseline, comparison group or independent evaluation is published.

A categorical training exclusion is published, with a cross-customer improvement claim beside it and no retention schedule. Artera says it does not use identifiable patient information to train its models. Its marketplace copy also says insights from its 1,000-plus customers continuously improve the platform, which the exclusion does not rule out if de-identified conversations are used; nothing states whether they are, what de-identification method applies to message text and call audio, or whether a customer can opt out.

A voice recording carries identity in the voice itself. Audio demonstrations are labeled as including real de-identified patient calls used with consent. No retention period is published for messages, transcripts or recordings. Safety engineering, meaning guardrails, hallucination mitigation and how a safety event is handled, is not described, although agents speak to patients directly and, in one customer's deployment, assess symptoms.

HIPAA status is stated and supported by published privacy material, with the agreement itself unpublished. Artera's federal and product pages state HIPAA compliance, and its trust center describes a privacy by design program, a privacy management system aligned to ISO 27701, and customer data hosted on Artera's own servers in onshore data centers; it also says identifiable patient information is not used to train its models.

The fuller documents sit in its SafeBase portal and privacy policy. No business associate agreement, its terms or the subcontractors it binds are published. Texts and calls pass through telecommunications carriers and messaging aggregators on the way to patients, so those parties belong in any agreement.

Artera names a broad set of attestations of its own, and the reports sit behind its trust portal. Company announcements list SOC 2 Type 2 covering security, availability and privacy (first issued October 2023), HITRUST certification held since 2019, and ISO 27001, 27017, 27018 and 27701; the trust center quotes Coalfire on the ISO management system.

Artera Harmony Federal Edition holds a FedRAMP High (Class D) certification with an agency authority to operate, which covers the federal edition rather than the commercial platform. The trust center page shows badges and links to a SafeBase portal for the documents. It does not state whether reports open without a request, the HITRUST assessment type or the report periods, which the portal itself may cover.

The company's stated position and one of its own customer stories point in opposite directions. Artera's pages say its AI agents automate administrative workflows and do not provide medical advice, diagnosis or treatment, with clinical decisions remaining with licensed providers.

The home page's customer story for a multi-location gastroenterology practice describes the agent platform performing intelligent symptom-based triage that identifies alarm symptoms and escalates to on-call clinical staff, and credits it with enhanced patient safety through triage that routes urgent symptoms. Sorting callers by their symptoms and routing the urgent ones to a clinician is triage, which the stated position rules out, and no regulatory position is given for it. A named, published triage protocol with a clinician's disposition behind each routing decision, or a stated regulatory position that matches this capability, would change the reading.

Multilingual reach is sold as a strength, and nothing is published on how the agents perform across languages. Artera's own marketplace copy says Harmony communicates in 109-plus languages with AI and LLM-driven translation, and a home page customer story credits its voice agent with natural conversation across multiple languages.

No per-language comprehension, translation or speech recognition result is published, and nothing describes a governance framework, bias evaluation or model review process; the solutions page says only that its AI monitors, improves and tests itself and flags upgrades for human review when appropriate. A caller the agent misunderstands, or an instruction translated wrongly, leaves no complaint and shows up only as a missed appointment. Artera does not claim equal performance across languages.

Mechanisms let staff check and correct an agent, and nothing stands behind a wrong outcome. Staff can take over any conversation with its full history, agents route complex requests to people, and every conversation is logged. The home page says results may vary and individual results are not guaranteed. No warranty, remedy, service credit tied to agent outcomes or measured error rate is published, and the one published rate, 65 percent AI agent task completion, carries no definition. The cost of an error falls on patients and the practice: a missed or double-booked appointment, a wrong balance quoted on a billing call, or an urgent symptom routed late in a triage deployment.

Named integrations with major record systems, documented by one of those systems and running in production. Artera's integrations page names Epic, Oracle, MEDITECH, Veradigm, athenahealth, eClinicalWorks, Greenway and NextGen, among others, through API, FHIR and HL7v2 connections that carry communication to and from the record, including updating appointment status.

MEDITECH sells the integration as its own Expanse Patient Connect product, integrated with its MyHealth portal, and publishes a hospital case study on reduced no-shows. The federal edition integrates with VistA, RPMS and Oracle Health. Nothing published says, for any one system, which events trigger a workflow, what an agent writes back, or whether it lands in a discrete field or as a note.

Onshore hosting is stated and a separate federal edition exists, while tenancy and retention are not published. Artera says all customer data is hosted on its own servers in onshore data centers holding SOC 2 attestations, accessed through the browser or its app. Artera Harmony Federal Edition runs under a FedRAMP High authorization, which constrains where that system runs and who administers it.

There is nothing on regions, the tenancy model separating one provider organization from another, where language model processing and translation run, retention for messages and call recordings, or the carriers and aggregators that carry texts and calls.

No price is published, and the sale is described. Artera sells an enterprise subscription to provider organizations, and to federal agencies through a General Services Administration Multiple Award Schedule contract. Its solutions page says custom agents built by its AI Service Squads, such as password resets, lab result notifications, prescription refills and insurance verification, are all included in the subscription. No figure, unit of charge (per message, conversation, provider or location) or implementation fee is published, and nothing says whether pricing depends on volume or outcomes.

Coverage is named widely, with validation behind part of it. Artera sells to specialty practices (behavioral health, cardiology, dermatology, gastroenterology, oncology, ophthalmology, optometry, ENT, orthopedics, pediatrics, primary care, urology and women's health), federally qualified health centers, health systems and federal agencies, the Veterans Affairs, Defense and Indian Health Service sites served through Artera Government Solutions.

It reports more than 1,000 organizations and over 2 billion patient interactions a year. Validation sits behind a few deployments: Atlantic Health's colonoscopy outreach with a dated result, a MEDITECH hospital case on no-shows, and several unnamed practice stories. The specialty pages describe workflows rather than results, and the 109-plus languages carry no per-language evidence.

The work centers on patient access within healthcare administration automation: agents that answer calls and schedule, reschedule and cancel appointments, collect intake forms and medical history, verify insurance and run referral outreach, with prior authorization listed among the workflows.