Parlance

Speech recognition and intent understanding are the engine of the product, delivered inside a managed service that carries much of the value. Callers say what they need and the system recognizes physician names, departments, specialties and requests, routes the call, books and manages appointments, reports prescription and referral status, and hands complex calls to staff with a summary.

The July 2026 Parlance 12 release applies statistical and large language modeling to first pass recognition and passes unrecognized intent to a large language model. The company's own commentary says healthcare voice vendors build on shared core technologies from large providers and that implementation and integration are what set them apart, and a health system buys hosting, directory management from HL7 feeds, call flow design, tuning and reporting alongside the recognition. Without the models there is no voice product, but a sizeable part of what is bought is the service around them.

The oversight structure is described in useful detail, and the company also promises its routing is always right. The structure: routine calls (scheduling, directions, prescription and referral status, routing, after-hours) are automated, while complex care coordination, escalations, sensitive situations and billing are listed as the health system team's work; transferred calls reach an agent with a verified patient and an AI-generated call summary; the switchboard supports emergency escalation and on-call routing protocols; the health system sets scheduling guidelines, escalation rules and exceptions, and Parlance tests and deploys flow changes and reviews interaction data continuously.

The promise: the switchboard page is headed 'Every Call Answered and Routed Correctly' and says the system 'routes the call correctly the first time'. A buyer who believes every call is routed correctly will not build the check that catches the caller sent to the wrong place, and since Parlance 12 an unrecognized request goes to a large language model rather than to an operator, so a misrouted call can leave no trace. The home page's 'Reach the correct person every time' is addressed to the caller and recorded with it.

Parlance describes its architecture in general terms and identifies nothing. Parlance 12, released in July 2026, applies statistical and large language modeling to first-pass speech recognition, passes caller requests outside the recognition vocabulary to a large language model that decides where to route them, and adds a modernized synthesized voice; a companion release adds AI-generated call summaries on transfer.

The company's commentary says healthcare voice vendors build on core technologies from large providers such as Google and Amazon. No provider or model is named for Parlance's own recognition, routing or summaries, and no accuracy figure, evaluation method, versioning or update practice is published.

The architecture is described and no provider is named. Since the July 2026 Parlance 12 release, caller speech outside the recognition vocabulary is passed to a large language model that decides where to route the call, first-pass recognition uses statistical and large language modeling, and transfers carry an AI-generated call summary.

Nothing names the speech recognition, synthesis or language model providers behind those steps, and the company's commentary that the segment relies on core technologies from providers such as Google and Amazon describes the industry rather than its own chain. The IT help desk assistant adds Cisco Duo for identity verification and ServiceNow as the ticketing system. No subprocessor list or statement of the terms that govern call content once it reaches an outside model is published.

Named customers and vendor-reported figures with no method. HCA Healthcare is described as running Parlance across more than 75 hospitals with over 70 percent of calls handled automatically, UW Medicine at nearly 4 million calls a year, Virtua Health as doubling annual call volume from 200,000 to 400,000 without adding contact center staff, and an unnamed $3 billion New Jersey health system as saving about $1 million.

Customer quotes add The Valley Health System answering up to half its incoming calls and an HCA regional CIO reporting return within 90 days. The headline figures (more than 85 percent of routine calls automated, more than 40 percent agent time saved, 87 percent caller engagement, 30 days to return) carry no baseline, denominator or period, and the company reports close to 2 billion calls answered across 400-plus health systems, which is scale of use rather than evidence of benefit. A contained call is not necessarily a correctly handled one. No independent evaluation is published.

Security assurances are published; the questions AI raises are not answered. Data is encrypted at rest and in transit, access is role-based and audited, and a breach notification process is in place, all within the business associate agreement.

Nothing published states whether calls are recorded, how long audio, transcripts or the AI-generated call summaries are kept, whether recordings improve the recognition or routing models, or what reaches the large language model that handles unrecognized requests.

One product feature carries a disclosure decision: patient room routing runs on record system feeds, so a caller who names a patient can be connected to that patient's room, and confirming someone is an inpatient is governed by directory opt-out rules. Nothing says how the opt-out flag reaches the routing layer or what the system says when a patient has opted out.

Business associate status is stated and the handling of protected health information is described. The switchboard page's FAQ says the product accesses protected health information through record system integration for patient room requests, that business associate agreement terms define data handling responsibilities, breach notification timelines and safeguards for the engagement, and that controls include encryption in transit and at rest, role-based access, consistent auditing and an incident response and breach notification process aligned with HIPAA.

The published security policy adds encryption at rest, TLS 1.2 or IPsec in transit, and annual third-party penetration testing. The website privacy policy covers the website only. The agreement itself is not published, so its terms and whether it binds subprocessors, including the large language model used for semantic routing, cannot be read before contracting.

A SOC 2 certification is claimed as the company's own, without its type or artifact. The switchboard page's FAQ says Parlance 'is HIPAA compliant and has SOC 2 certification', with encryption in transit and at rest, role-based access, auditing and an incident response and breach notification process.

The published security policy describes the controls behind it: least-privilege access, customer data encrypted at rest, TLS 1.2 or IPsec in transit, third-party external penetration testing at least annually with findings remediated within a service level, internal vulnerability scanning, centrally managed endpoints and monthly staff security training. The SOC 2 report type (Type 1 or Type 2), its period and scope are not stated, and there is no trust center or request route for the report.

No device claim is made and the core work is administrative: routing calls, booking appointments, reporting prescription and referral status, answering common questions and directory lookups. One function sits near the clinical boundary and is not described: the switchboard page lists 'emergency escalation protocols' and 'emergency and on-call routing protocols' among the workflows it supports, and a hospital switchboard receives callers describing emergencies in the same queue as routine requests.

Nothing says whether the system recognizes an emergency from what a caller says, what it does then, or whether a clinician is involved. Outbound automated calls and texts carry telephone consumer protection consent obligations, and UK deployments sit under UK data protection law.

Behavior is monitored and changes are tested as part of the managed service; nothing is published as evaluation. The contact center page says the Parlance team continuously reviews interaction data to improve containment and routing, tests and deploys every conversation-flow change the health system requests, and holds regular performance reviews against the customer's goals.

The July 2026 release reports a significant improvement in first-pass recognition accuracy, so accuracy is measured, but no figure, method or break-down by caller group is published, and no governance framework or bias assessment is described. Recognition quality is marketed as a strength ('the accuracy required for healthcare call routing'), and recognition varies with accent, first language, age, hearing and distress, which are common among people calling a hospital. A caller the system does not understand leaves no complaint, so a gap by group stays invisible without measurement.

Mechanisms exist to review what the system did, and nothing stands behind its results. The managed service includes continuous review of interaction data, regular performance reviews against the customer's goals, service level agreements with defined escalation procedures and rapid response for issues affecting patient access, and transfers that hand the agent a call summary. Those answer for availability and support rather than for a wrong route or a wrong answer.

The containment and automation figures carry no method, and no misroute rate, warranty, remedy or credit attached to a call the system handles wrongly is published. The switchboard's promise that every call is routed correctly has nothing standing behind it that a health system could hold the vendor to.

Integration is standards-based with named deployments behind it; no record system is named. The platform reads live provider schedules so availability is current when a caller asks, writes bookings and changes back, surfaces prescription and referral status, populates the agent's desktop with a verified patient and call summary on transfer, and updates provider and department directories from HL7 feeds every fifteen minutes; patient room routing also runs on HL7 feeds from the record system.

The company says it integrates with all major record systems and that its customer solutions team includes people from record system vendors, and deployments at HCA Healthcare, UW Medicine and other named systems stand behind the claim. It sits alongside existing enterprise telephony, and the help desk assistant is on the ServiceNow Store. No record system is named, no marketplace listing is published, and no FHIR position is stated.

Deployment options are stated; cloud residency and the processing path are not. The switchboard FAQ describes on-premises deployments with redundant service units in the customer's environment and service levels matched to the customer, and hybrid or cloud deployments where customers extend their telephony across multiple geographic zones with failover back into their own environment if the internet fails. Under the managed service the vendor hosts, monitors and updates the platform.

An on-premises deployment keeps call handling inside the health system's environment, but nothing says where cloud deployments run, which regions UK and Canadian customers use, or where the large language model behind semantic routing and the call summaries is hosted, even for on-premises customers.

No price, rate card or unit of charge is published, and the shape of the offer is discoverable. Parlance is sold as a fully managed service (the vendor hosts, monitors, tunes and updates the platform under service levels) across switchboard, patient access contact center, clinic and call routing products, a replacement for Nuance Speech Attendant, an IT help desk assistant on the ServiceNow Store, and SMS and web chat, in on-premises, hybrid or cloud deployments, to health systems in the United States, Canada and the United Kingdom as well as enterprise and higher education.

It states an average time to return of 30 days and implementation in six to eight weeks. The company is owned by Constellation Software through Volaris Group, a buy-and-hold owner, which bears on continuity rather than price.

The product automates the health system's phone front door. It answers switchboard and patient access calls for appointment booking and changes, prescription refill requests and status, referral status, directory and department routing and after hours coverage, and since September 2026 it takes IT help desk calls through the ServiceNow Store, including password resets with voice identity verification.

Coverage by organization type is broad and named: a national hospital operator across more than 75 hospitals (HCA Healthcare), academic medical centers (UW Medicine, UC San Diego Health, NYU Langone Health), children's hospitals, community and regional systems, and clinics, across the United States, Canada and the United Kingdom, by voice, SMS and web chat. Validation sits behind part of it through named customers and figures. No language coverage is published, which matters for a product answering the front door of urban and safety-net systems, and every buyer named is a provider organization.