Vi
Vi sells what it calls an AI execution layer to health systems, payers, life sciences companies and wellness businesses. Its models decide the next step for a patient, member, prescriber or operations team and trigger it inside systems the customer already runs. Vi Activate predicts who is likely to start care, enroll in a program or qualify for a trial, and aims outreach at them. Vi Engage picks who to contact, on which channel and when, testing variations against each other and retraining every week. Vi Operate, the agent suite launched in May 2026, covers patient communication, enrollment scheduling, staffing, operating room scheduling, supply chain and market research. Vi Pulse is the command center for querying the data and tracking return on each deployment.
Underneath sits the Vi Data Web, which joins a customer's de identified clinical, claims and operational data with licensed consumer and behavioral data the company says covers 96% of US households. That data is the core of the pitch. The company's website privacy policy also says it sells and shares personal information for targeted advertising, including custom audiences it supplies to customers.
Sales run through a demo and a three month proof of concept, after three to four weeks of setup, committed to a 4X return with 1X downside protection measured against control groups. Vi reports more than 100 enterprise customers, names Minnesota Mental Health Clinics among them, and raised $145 million at a $1.64 billion valuation in May 2026. It is based in New York and Tel Aviv. The trust center lists ISO 27001 and ISO 27701 certificates, a SOC 2 Type 2 report audited by PwC and a public subprocessor list, with HITRUST in progress.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Prediction is what Vi sells. Activate scores which patients, members or trial candidates are likely to act, Engage chooses the person, channel, message and moment for each contact and retrains every week on what happened, and the Operate agents carry the next step into scheduling, staffing, supply and outreach work. Pulse puts a language model interface over the data and the return figures.
The licensed consumer and behavioral data in the Vi Data Web feeds those models rather than being sold beside them as a separate product, and the commercial offer itself, a return measured against a control group, is a promise about what the models produce. Take the models away and what remains is a data warehouse and a dashboard.
The agents are built to act without waiting. Operate's agents recommend, execute and improve workflows without manual effort, and the outreach agents answer questions and book appointments over text, voice and email.
Oversight comes as phrases rather than mechanisms: operators stay in control, decisions requiring judgment escalate to a person, outreach on the life sciences side has approval built in, and every action is traceable. Which decisions count as requiring judgment, what an approval step covers, the point at which a case passes to staff and what happens when an agent books or sends the wrong thing are not described.
The engagement engine is described in working terms. Propensity models predict who will respond and when, multi armed bandit tests choose among channels, content and timing, and the models retrain weekly on results. Operate builds a live digital twin of a customer's operations from its data before agents act, and Pulse uses large language models to search data and write reports.
The language models are not identified. Vi calls itself model agnostic, bringing newly released models from several providers into its products and letting customers use third party models where they perform better, but it does not name the providers or versions behind any agent, or say how customers learn that a model has changed.
The subprocessor list, dated September 2025, is public and names four parties, all processing in the United States: Amazon Web Services for hosting and storage, Cloudflare for network delivery and security, Snowflake for data warehousing and Spatial.AI for data enrichment and geospatial audience analytics, the last flagged on the list itself as included if still core.
No language model provider appears, although agents and Pulse are described as running on models from outside providers. The licensed consumer and behavioral data feeding the Data Web comes from suppliers that are not named. Changes to the list follow a notice and objection process set in the data processing agreement.
Results come mostly from anonymous case studies: a multi site health system, a regional payer, a hospital system running operating room scheduling, a clinical stage biotech and a Fortune 500 biopharma company. Reported outcomes include $28 million in new revenue for a mental health clinic chain from enrollment scheduling, nine months cut from a trial timeline and a 19% net script lift against a control group.
One customer is named. Minnesota Mental Health Clinics grew from 45 clinicians seeing 500 patients a week to 150 clinicians seeing 2,500 a week in just over two years while using Vi for patient outreach. The control group comparisons come without sample sizes, periods or method, the headline figures (a 4X return, costs cut by up to 50%, trials 76% faster) carry no source, and no peer reviewed or independent evaluation is published.
De identification is the central commitment. The Data Web runs on de identified data, partner data is de identified before use, and the data processing agreement bars attempts to reidentify it. Customer personal data is kept for the life of the contract and deleted or returned at the end, at the customer's choice.
Training is where the commitments stop. Engage's models retrain every week on engagement results, and agents are described as trained on more than 190 million de identified psychographic and demographic profiles, so customer outcomes feed the models in de identified form with no stated limit on that use. The website privacy policy says the company sells and shares personal information for targeted advertising. Nothing is published on guardrails, hallucination controls or how a safety event in an agent conversation is handled.
HIPAA compliance is claimed on the healthcare pages, and the trust center lists HIPAA among its compliance items, yet no page states that the company acts as a business associate or offers an agreement. The data processing agreement anyone can read is written for GDPR, UK law and US state privacy laws, treats health data as sensitive data where a customer supplies it, and never mentions HIPAA.
Partner data is also said to be de identified before use. That statement does not reach the agents that send appointment reminders, schedule enrollments and answer patients by text, phone and email, work that needs identifiable contact details, and how patient information in those workflows is covered is not stated.
Three certifications are named in detail. ISO/IEC 27001:2022 was issued to Vi Labs Ltd by the Standards Institution of Israel with IQNet recognition in 2026, valid until 2029 under registration IL-1127137, for the IT operations behind its cloud AI platform for healthcare and life sciences. An ISO/IEC 27701:2019 privacy management certificate carries the same term, and a SOC 2 Type 2 report is audited by PwC. HITRUST is shown as in progress.
The certificates and the SOC 2 report sit behind an access request in the trust center, while the customer terms, data processing agreement and subprocessor list open to anyone. The SOC 2 report period is not given.
Everything Vi sells works on acquisition, engagement and operations: aiming outreach, timing reminders, booking enrollments, scheduling operating rooms and staff, forecasting supply and segmenting prescribers for drug marketing. Patient care navigation and next best actions are described as outreach and routing rather than clinical judgment, and patient matching for trials finds likely candidates in licensed EHR and claims data for sponsors to recruit.
No device claim is made, and none of the marketed work reaches diagnosis or treatment. No regulatory statement is published beyond that, which is the usual footing for commercial and operational software in health care.
Testing at Vi means measuring return: campaigns and agent actions are compared with control groups, and Pulse reports the lift. Governance otherwise appears as language. Actions are described as governed and traceable, the AI as applied responsibly, and a healthcare advisory committee of industry executives serves as a strategic sounding board.
No bias or fairness evaluation, results by group, AI management certification or model risk documentation is published. Targeting draws on psychographic and demographic profiles to choose who is contacted about mental health care, trials and prescriptions, so results by group would show who the models reach and who they pass over; none are reported.
Vi puts its commercial terms behind results. The proof of concept is committed to a 4X return with 1X downside protection, measured against control groups, and the healthcare pages repeat the promise for every dollar spent. Read plainly, a customer is protected from ending below what it paid, a testable commitment about what the models deliver in aggregate.
It does not attach to any single output. An outreach message sent to the wrong person, a wrong booking or a poor staffing recommendation carries no stated remedy, and nobody outside the customer has a route to correct one. Whether the protection comes as a refund, a credit or further work, and who runs the control group, are not spelled out in public terms.
Existing record systems, CRMs and data warehouses stay in place. The platform runs alongside them, with no rip and replace and no migration. A health system case describes several EHR systems unified through one outreach platform, and a payer case describes integration with the payer's existing omnichannel communication tools.
No EHR, CRM or messaging platform is named, no marketplace listing or interface standard is cited, and whether agents write back to the record or only read from it is not described. The trust center lists an integrations document, which sits behind an access request.
The platform runs on Amazon Web Services, and every party on the subprocessor list processes data in the United States. Enterprise data is anonymized as it is loaded, and the trust center describes an alternate processing and storage site with a four hour recovery time objective.
Vi also says its models deploy into a customer's existing infrastructure, without saying whether that means the customer's own cloud. How one customer's data is kept apart from another's and from the licensed consumer data in the Data Web, and which AWS regions hold it, are not stated.
The sales motion is published in detail and the price is not. After a demo, a customer's anonymized data is loaded in about three hours, the expected return is underwritten in a day, setup and model pre training take three to four weeks, and a three month proof of concept follows, committed to a 4X return with 1X downside protection measured against control groups. The company describes the arrangement as winning when its customers win.
The unit of charge after the proof of concept is not stated, whether per product, per agent, per outcome or by data volume, and neither is the price of Activate, Engage, Operate or Pulse on its own or how the downside protection is paid if the return falls short.
Four buyer groups are named: health systems and hospitals, payers, life sciences companies and wellness businesses such as health club chains. The use cases differ for each. Health systems get patient acquisition and reactivation, operating room scheduling, staffing and supply chain agents. Payers get member engagement and preventive care nudges, including a musculoskeletal program. Life sciences companies get trial candidate identification, site selection and prescriber targeting.
The trial recruitment and site selection work and the operating room and staffing agents each sit on their own solution page with a case study. Validation behind any setting is thin and mostly anonymous; the one named provider, Minnesota Mental Health Clinics, stands behind the behavioral health outreach work.
Alternatives to Vi
Other products that do the same kind of work as Vi, and what each one does.
Innovaccer sells an agentic platform for health care. Its Healthcare Autonomy Platform runs agents and copilots on Gravity, a data layer that unifies EHR, claims, pharmacy, laboratory and social determinants data across a reported 80 million patient lives.
Notable sells an AI agent platform for health system and health plan operations.
Qventus automates hospital operations.
QuantHealth is a clinical trial simulation company with offices in Tel Aviv and Worcester, Massachusetts.
GE HealthCare Command Center is hospital operations software that pulls streaming data from a hospital's record and resource systems into tiles, small applications that each cover one part of capacity, patient flow or care progression, from Capacity Expediter, Patient Manager and Hospital…
TeleTracking, based in Pittsburgh, sells patient flow and capacity management software to hospitals and health systems in the United States and the United Kingdom.
Compared With
Each comparison carries a written verdict, the buyer conditions that favor each vendor, and a graded side by side. Pairs that cross a category boundary are grouped separately, and their verdicts state where the boundary sits rather than manufacturing a head to head.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not published.
|
Vi is sold as an enterprise platform after a demo, starting with a proof of concept. Setup and model pre training take three to four weeks, then the proof of concept runs for three months, committed to a 4X return with 1X downside protection measured against control groups. The unit of charge after the proof of concept is not published. | No business associate agreement is published. The public data processing agreement covers GDPR, UK law and US state privacy laws and does not mention HIPAA. | Not published. | Vendor Published |
Vi sells its platform through a demo and a three month proof of concept tied to a measured return, then an enterprise contract. No price, tier or unit of charge is published for Activate, Engage, Operate or Pulse. Whether products and individual agents are priced separately, how the 1X downside protection is paid if the return falls short, and what the contract looks like after the proof of concept are not stated.