SpinSci Technologies

The models now drive the lead product, inside a platform that carries value of its own. SpinSci sells itself as agentic AI for patient access: voice agents that schedule, handle refills, billing, benefits, prior authorization and referrals end to end, and outbound agents for reminders and campaigns, all running on its Healthcare AI Fabric of a patient access knowledge graph, a reasoning engine and an unnamed language model. Those agents do not exist without the models.

Around them sit two decades of contact center integration that stand without AI: computer telephony integration with Epic, the agent desktop that surfaces patient data, an operator console and hospital clinical communications with on-call schedules.

The core promise is that the agents do not falter. The Voice AI page says the agents "handle every inbound patient interaction at any volume, at any hour, with no degradation in speed or accuracy", and the home page says scheduling, referrals, billing and prescription requests are resolved accurately, at any volume, around the clock. No escalation design fixes that promise: a buyer told accuracy never degrades will not build the review that would catch it.

The company's own terms say the opposite, that AI models may occasionally produce inaccurate or incomplete results. The oversight described is thin: agents complete tasks end to end without human escalation and escalate when a patient asks or "when the agent determines escalation is appropriate", with the live agent receiving full context; no criterion, threshold or hours of human availability are published. Answering every call at any hour takes on the caller in distress at 2am.

The approach is explained and no model is named. The AI Fabric page describes a patient access knowledge graph combining a domain ontology with record and journey data, a reasoning engine that plans multi-step interactions, semantic caching so conversations continue without calling the language model at every turn, and decision trees drawn from the health system's own record configuration.

Answers are described as grounded in record, payer, pharmacy and knowledge graph data, and the fabric as trained on 400 million patient interactions a year. No language model, supplier or version is named, and the terms say only that models are continuously improved and results may vary over time. The 98 percent natural language success rate has no definition or method.

The architecture is described and no model provider is named. The AI Fabric page places an unnamed language model behind a reasoning engine and semantic cache, and says the fabric runs AI agents from SpinSci, its partners or the customer, so other companies' agents may act through it on the record.

The parties named are integration partners: Epic and Oracle Health on the record side, and Cisco, Zoom, NICE and Avaya on the contact center side, with Webex deployments reaching the customer's record system through a site-to-site VPN. The privacy policy names only categories of website service providers. No hosting provider, model provider or subprocessor list is published.

Scale and customer quotes, with no method behind the figures. The home page reports 165 health systems, more than 400 million patient interactions and 60 million patients served, and quotes staff at unnamed health systems with their location and provider counts: 300,000 minutes of agent time saved by the Transfer Center, 40 to 50 seconds saved per automated patient verification, 30 seconds saved per call across just under 850,000 engagements in 2025, and a fivefold increase in proactive outreach.

The Voice AI page adds a 98 percent natural language success rate. None names the organization, baseline or measurement method, most of the figures describe the integration and agent desktop products rather than the voice agents, and scale of use is not evidence of benefit.

Two published statements pull against each other. The AI Fabric page says "PHI stays inside your perimeter", while the same page says the fabric is trained on 400 million real-life patient interactions a year and that the agents learn from millions of patient interactions across health systems. The terms add that users keep ownership of their health data but grant a license to use it to provide and improve the services.

The privacy policy keeps personal information as long as reasonably necessary and addresses website data, not call content. Nothing states whether call recordings and transcripts are retained, whether interactions are de-identified before training, or whether one health system's calls improve agents serving another.

HIPAA compliance is claimed and nothing is published behind it. The AI Fabric page lists HIPAA among compliance built into the architecture and says patient information is safeguarded by limiting access and honoring consent. The privacy policy covers the website and related services but addresses only contact, form and cookie data, and no business associate agreement or service privacy notice is published.

Business associate status is structural: the agents read and write the record, verify identity and take payments. The contracting boundary is layered, since a Webex deployment involves agreements with Cisco for the contact center, SpinSci for the integration and agents, and the record system vendor, with patient data moving across all three.

The company presents SOC 2 Type II as its own. The AI Fabric page lists HIPAA, SOC 2 Type II and PCI under compliance built into the architecture, and says the company meets industry standards through regular audits and transparent reporting. No period, auditor, scope or report is offered, and the site links no trust center or security page. Cisco's own HITRUST and Webex health privacy statements cover the contact center platform the product runs inside, not SpinSci's integration layer.

No regulatory position is stated, and the company's terms and its product marketing describe the product differently. The terms published with the privacy policy say AI-generated insights "are for informational purposes only and do not constitute medical advice" and that the platform "is a tool to assist healthcare providers, not replace them".

The Voice AI page says the agents give "responses to common clinical questions based on approved content" and that these tasks are "completed end-to-end without human escalation"; the pharmacy agent helps patients with medication questions. The answer goes to a patient, not to a clinician, and the approved content is the health system's, unpublished, with the agent composing answers from it.

The company's own guides set some limits: its patient access guide says the agent stops and routes to a person when a caller mentions something like chest pain, and its page on hold times says complex clinical questions go to a live agent. Those limits do not reach the common questions answered without escalation. Outbound voice campaigns also carry Telephone Consumer Protection Act consent duties for AI voices.

Governance is described as features, with no evaluation behind them. The AI Fabric page says the platform embeds guardrails, system reliability and observability into every workflow, and that answers are grounded in record, payer and pharmacy data. No governance framework, evaluation method, error rate by task or subgroup result is published. A voice front door depends on the agents understanding every caller equally.

Speech recognition degrades with accent, second language, age and speech impairment, and a caller the agent fails to understand may hang up and leave no record of trying to book, refill or report a symptom. The site markets no multilingual operation and gives no result by language or accent.

Mechanisms exist to check an output, and the published terms stand behind none of it. The platform describes observability across every workflow, responses grounded in record, payer and pharmacy data, and handoffs in which the live agent receives the full conversation, so a disputed interaction can in principle be reconstructed.

The terms published with the privacy policy say AI models may occasionally produce inaccurate or incomplete results, exclude liability for errors or omissions in AI-generated content, and cap aggregate liability at twelve months of fees. They read as template text written for a consumer health app rather than for a contact center platform. The person harmed by a wrong booking, a missed refill or an unanswered symptom is a patient with no route to the vendor.

Named bidirectional integration with major record systems, verifiable outside the company's own claims. SpinSci states bidirectional integration with Epic and Oracle Health, reading live patient data and writing updates back, and runs natively with Cisco, Zoom, NICE and Avaya contact center platforms.

The Cisco Webex App Hub lists its Patient Assist integration with the requirements to enable it, including a site-to-site VPN into the network where the record system is hosted; Intermedia's own support documentation describes its contact center integration through SpinSci with HL7 record systems including Epic, Cerner and MEDITECH; and the company announced an Epic App Orchard listing in 2019.

A health system executive on the home page credits SpinSci with standing up its computer telephony integration with Epic, which is production evidence. What remains open is currency. The Epic listing was announced in 2019, and the company does not say which record systems the voice agents write to today.

The placement is stated and the processing path is left open. The AI Fabric page says protected health information stays inside the customer's perimeter, and the Cisco Webex App Hub listing says the integration reaches the record system through a site-to-site VPN into the customer's own network rather than by extracting records to a vendor estate.

What is not described is where SpinSci's own components run: the language model behind the reasoning engine, the semantic cache that keeps track of prior interactions, the knowledge graph holding a longitudinal view of each patient, and the training said to draw on interactions across health systems. No hosting provider, region or tenancy model is published.

No price is published. The product is sold by demo, and the Cisco Webex App Hub listing shows a second route: SpinSci licenses ordered through Cisco's commerce system alongside paid Webex Contact Center licenses, with pricing on Cisco's commerce site. The company does not say which unit it charges by, whether per agent seat, per interaction, per completed task or platform subscription. The integration products are priced by seat, while voice agents priced per interaction scale with call volume. Whether buying through Cisco changes the terms is not stated either.

Coverage is named, with validation behind part of it. The buyer is the health system patient access organization and contact center. The voice agents cover scheduling, pharmacy and refills, billing and payments, benefits verification, prior authorization and referrals, with outbound campaigns for reminders and no-shows; alongside them sit the agent desktop, transfer center, operator console and hospital clinical communications.

The integration and contact center products are supported by quoted health system staff with location and provider counts; the voice agents' coverage rests on the company's description and one quoted figure. It is not a clinical product, though the agents answer common clinical questions from approved content.