Relatient

The company's own framing is rules driven automation, and that is the honest reading. What makes scheduling work here is an engine encoding each provider's preferences and constraints, built over years, and that engine would function with no model at all through the online booking interface it also powers.

The voice agent is genuinely conversational artificial intelligence and it is a channel onto that engine rather than the thing being sold. The July 2026 clinical request capture is a further model dependent capability, since extracting a request from unstructured speech and routing it is not a rules problem.

So this is a scheduling platform with real artificial intelligence attached, graded the way this index grades a practice management system with a scribe or an intake platform with a voice product.

High autonomy on scheduling with the boundary drawn in a defensible place. The agent books, cancels, reschedules and confirms with what the company calls zero staff intervention, and writes the result back to the source system without a person checking it.

What it does not do is decide anything clinical. The clinical request capability captures what the patient asks for and creates a task for the care team; it does not answer, triage or prioritise.

The oversight mechanism is unusual and worth crediting: the agent is constrained by the same rules engine the human schedulers use, so the limits on what it can do are the organisation's own configuration rather than a model's confidence threshold. Held at B because nothing published states what share of calls complete without transfer, and because no urgency detection is described. A patient describing a worrying symptom to a scheduling agent is a foreseeable event, and a competing vendor in this index publishes explicit stop logic for exactly that case while this one does not.

The workflow is described clearly and the technology is not. Public material explains the rules engine, the shared logic between human and automated scheduling, the real time write back and the handoff to text based self scheduling.

No model is named, no containment or transfer rate published, no accuracy figure given for the clinical request extraction, and no statement of which languages the agent handles. For a product answering a large share of a health system's inbound calls, the transfer rate is the number an operations leader would work from and it is absent.

Two external assessments sit behind this rather than a claim, with both a healthcare specific certification and an audited service attestation stated alongside compliance and single sign on. That is better than the category norm and it establishes that someone outside examined the controls. What a buyer should notice is that the data category has changed under the same product name.

A scheduling platform sounds like it handles appointment slots, and this one records or transcribes calls to extract a request, and a 2026 capability means clinical and medication content now flows through the voice channel. Those are different kinds of information governed by different expectations, and a health system that assessed this vendor as a scheduling tool assessed a narrower product than the one now deployed.

Scope creep of that kind is invisible unless someone re reads the description, and it is worth adding to any renewal review. Nothing published states a retention schedule, a recording policy, or whether call audio contributes to model development, and no model, hosting arrangement or sub processor list was located, which matters in voice because a call passes through several tiers before reaching a model. Ask what the voice channel now captures, whether calls are recorded and for how long, whether audio trains models, and for a sub processor list covering the voice path specifically.

Independent recognition plus named customers with figures, and no study.

Best in KLAS in 2024 for patient self scheduling and engagement is a third party ranking drawn from provider interviews, which this index treats as real evidence of delivered value rather than marketing. Scale corroborates it at more than 50,000 providers and over 150 million appointments a year.

Customer results are named rather than anonymous, including a primary care organisation reported to generate more than 60,000 self scheduled appointments a month. Held at B because those are volume figures rather than outcomes: nothing published measures no show rates, time to appointment, abandoned call rates or access equity against a baseline.

Better than the category norm because the controls are externally assessed rather than asserted, with HITRUST and SOC 2 both stated alongside health privacy compliance and single sign on.

The data handled is broader than a scheduling platform sounds. Calls are recorded or transcribed to extract a request, and the July 2026 capability means clinical and medication content now flows through the voice channel, which is a different category of information from an appointment slot. No retention schedule, recording policy or position on whether call audio contributes to model development was located.

Rests on assessed controls rather than a claim, with HITRUST certification mapping a control set onto health privacy requirements and SOC 2 alongside it.

No agreement posture was located publicly. At more than 50,000 providers the agreements are plainly standardised, and the specific term to establish is how call recordings and transcripts are treated, since those are the artefacts this product creates that a scheduling system historically did not.

Two independent certifications named, HITRUST and SOC 2, which puts this above the many vendors here that assert compliance without assessment.

What is missing is the tier, and this index has repeatedly shown that the tier is the information. HITRUST has three assessment levels of very different demands, and SOC 2 reports come as a point in time Type 1 or a period of operation Type 2. Another vendor built in this sweep publishes its equivalents as r2, Type 2 and Level 1, and that specificity is exactly what separates an A from a B here. Held at B, and a buyer should simply ask which.

No device pathway applies and none is claimed. Booking an appointment is administrative.

The boundary worth watching is the one the company crossed in July 2026. Capturing a clinical or medication request by voice and routing it to a care team is message handling, not clinical judgement, and it stays outside device regulation on that basis. It would move closer if the system began prioritising those requests by clinical urgency, because ranking patients by how sick they sound is a different object from creating a task. Nothing suggests it does; a buyer should confirm that it does not.

Automated outbound calling also sits under telecommunications rules on artificial voice, an exposure this index has recorded on other patient outreach products.

Nothing published on evaluation, monitoring, error rates or performance variation, and this product has the broadest exposure of any voice record in this index because it answers the general access line for more than 50,000 providers.

Automated speech recognition performs unevenly across accent, dialect, age and speech difference. When the system handling those calls is the front door to an appointment, a caller the model repeatedly fails to understand is a caller who does not get booked, or who is transferred into the queue the automation exists to shorten. That is access, not convenience.

This is the sixth instance of the speech equity exposure recorded in this index and the largest by call volume. No subgroup performance, no transfer rate by caller characteristic, and no statement of language coverage were located.

One design property here is a genuine consistency guarantee and it is easy to overlook. The public material explains that the automated and human scheduling paths share the same logic, alongside a described rules engine, real time write back and a handoff to text based self scheduling.

Shared logic means the agent applies the same rules a staff member would, so a patient does not get a different answer depending on which channel they used, and a scheduling error is a rule error rather than an agent error. That is a better structural position than a separate automated path tuned for throughput, and it makes a disputed booking traceable to a rule someone wrote. Held at C because nothing is measured and one figure in particular is absent.

No transfer rate is published, and for a product answering a large share of a health system's inbound calls the transfer rate is the number an operations leader actually works from, since it determines staffing and it is the honest measure of how much the system does. No accuracy figure for clinical request extraction, no containment rate, no statement of which languages the agent handles, and no warranty, indemnity or remediation commitment was located. The language question is not incidental for a patient facing line. Ask for the transfer rate, extraction accuracy, the language list, and what happens when a caller describes something urgent.

Among the stronger integration positions in this category. Named connections span the major record and practice management systems, including Epic, Oracle Health, athenahealth, NextGen, Veradigm and ModMed, over application interfaces and health data exchange standards, and the platform is described as working across mixed record estates, which is the situation most multi site groups are actually in.

The write back is the operative part: appointments post to the source system in real time, and the newer clinical request capability creates a task inside the workflow tool the care team already uses rather than in a separate queue. Held at B because no certification level or interface detail is published and third party guidance puts a typical integration led deployment at around 90 days.

A hosted platform with single sign on and role based access, deployed against the customer's record estate. Third party guidance describes integration led deployments taking roughly 90 days.

No hosting model, region or retention position was located, and the retention question matters more here than for a scheduling system alone because the voice channel creates recordings and transcripts.

The mechanism is described by third party guidance even though no rate is published anywhere: pricing is quote based and modular, varying with the number of providers and locations and with messaging volume.

Knowing that messaging volume is a variable is useful, because it means the cost moves with patient contact rather than with headcount, and an organisation that automates more communication pays more for doing so. A buyer should establish whether the voice agent is priced within the platform or as a module, what the unit is for voice specifically, whether it is per call attempted or per call completed, and how the new clinical request capability is charged, since it was added years after the platform and addresses a different budget.

Broad and enterprise oriented. Buyers span hospitals and health systems, multi specialty groups and physician practice management organisations, with the platform positioned for complex provider rules, high call volume and limited after hours coverage.

Specialty coverage is handled through the rules engine rather than through specialty specific models, which is a real design difference from the competing voice vendors in this index that train per specialty. Independent guidance names orthopaedics, gastroenterology and dermatology as particular fits, and the reach of more than 50,000 providers and 150 million appointments a year is among the largest in the category. Held at B because everything sits at patient access, with nothing extending into the clinical encounter, and coverage is United States only.