Inova, a large nonprofit health system in Northern Virginia and Washington, DC, that serves more than 4 million patient visits annually across hospitals, primary care clinics, and urgent care centers, has evaluated more than 300 AI-enabled digital health applications and deployed more than 70 different AI apps.
To support the use of AI, the health system recently implemented a data modernization effort in collaboration with data integration company Fivetran. Inova accelerated its four-year data modernization roadmap to six months to enable AI and move implementation from experimental to operational stage.
To operationalize AI, the health system needed to align several technology platforms, including Fivetran and DBT, which completed their merger on June 1. Overall, the health system manages more than 1,000 technology contracts. Other partners include Astronomer, Microsoft Fabric, and Databricks, which supports advanced analytics and AI workloads for healthcare systems.
The health system’s software platform consists of clinical, imaging, laboratory, financial, and Software-as-a-Service solutions. According to the Fivetran case study, each of these platforms has its own data model, application programming interface, and limitations.
Inova has also developed its own digital AI delivery portal with an Inova branded look and feel. Inova staff will be able to access AI agents from other vendor platforms on the health system’s portal.
Developing an AI application strategy with Inova
John McManus, Inova’s chief data and AI officer, said the health system is not trying to compete with third-party app developers in deploying software across the organization. It just incorporates some of these AI features.
“By centralizing and standardizing data movement with Fivetran, Inova has created an open foundation to support AI agents, co-pilots, and intelligent applications built on trusted, controlled data,” the case study states.
Inova’s third-party apps include an app that uses AI algorithms to search for microcoagulation in post-stroke chest CT scans. In addition, Inova uses the latest models from OpenAI and Anthropic and is developing smaller models and features based on those models, McManus said.
“In some cases you may want to control the UI [user interface]the look and feel,” he explained.
Additionally, health systems will teach their employees how to become AI-savvy and get them used to using some AI features as part of their workflow. McManus likens interacting with these AI features to using a directions app at a complex theme park like Disney.
“We need a map. We need to know where the restrooms are. We need to know if there’s an app to book a ride and what its status is,” McManus said. “We need a similar mindset in healthcare: How can we provide a pathway for people to become AI-savvy with workflows that deliver these AI capabilities?”
He added that it is difficult for medical professionals to become proficient in AI if they have to work with “100 different representations of AI.”
On the customer-facing side, Inova’s more than 1,000 providers use Abridge as an ambient AI scribe, mostly in outpatient or emergency settings, McManus said. McManus said the health system is piloting AI ambient scribes in some specialty inpatient areas, using agent AI to refer patients to health care providers such as dermatologists who are not part of the Inova network.
These AI voice agents send data back to Inova, allowing Inova’s clinicians to assist community partners if a patient develops a complication, McManus said.
Another key AI feature of Inova is to prompt healthcare providers to ask questions of patients, reducing the burden of paperwork and approvals. Inova is developing additional apps with similar functionality. For example, Inova is working with health insurers Elevance Health and Abridge on a pilot project to reduce the administrative burden of pre-charting workflows, McManus said.
The project includes Inova sharing the names of Elevance members with upcoming appointments with providers. We then send data about these members to Abridge so that providers know what to cover on the exam while using Ambient Scribe.
“Ensuring additional information is recorded in the notes will allow for faster approval from payers for services ordered for medical care,” McManus said.
Inova is also developing AI-driven capabilities that operate in the background and do not interact directly with patients. For example, McManus said, the company is developing an AI tool that searches for available provider schedules and distributes that data to health systems’ access teams, so they can “proactively” book patients.
However, Inova needed to implement guardrails when building an IT architecture to support these AI tools.
Avoiding vendor lock-in and setting guardrails
As Inova evaluates data modernization partners to enable AI, health systems want to avoid being locked into vendor contracts, McManus explained.
“We choose our partners carefully,” McManus said. “We try to avoid being too single-threaded with the underlying techniques and technologies that they provide.”
He explained that the importance of incorporating diverse products into the technology stack rather than working with a single vendor or model developer means that it is important for health systems to have the flexibility to operate and change their IT infrastructure while protecting the organization’s core business logic (referring to the unique rules and algorithms that act like the “brain” of a software program).
Inova turned to DBT Labs to create a shared data infrastructure layer that supports data quality and governance. This layer has allowed health systems to improve the reliability of data across the organization.
McManus also said that Fivetran has enabled Inova to quickly scale its data infrastructure. Health systems have standardized data movement using a single, managed ingestion model, which is how data is collected and moved to storage. Inova used Fivetran’s Connector SDK to extend its uptake model across the healthcare system.
However, he emphasized that healthcare organizations need consistent goals and architectural designs when working with multiple technology partners. Not only that, but healthcare organizations must also create guardrails for AI apps to protect patient privacy.
For example, with Inova, once a patient enters the exam room, the provider uses AI to have a conversation with the patient about the recorded visit, ensuring the data is not mishandled, McManus noted.
He further explained that investing in the right vendors to build these guardrails can protect not only patients but also health system operations. He said the guardrails will allow AI to be deployed responsibly and at scale, following the “mechanisms” of health system governance and policy.
Ensuring that AI apps remain compliant and adhere to health system policies is a key focus, McManus emphasized. This means the health system holds all application vendors to the same validation standards.
“Every pipeline, every agent, every model is judged on whether it makes health care safer, more equitable, or more accessible,” he said. “Governance doesn’t have to be a barrier to AI, but is actually a prerequisite for success. So we’re weaving deeply into these frameworks to implement our solutions, and governance in a way that maintains reliability, performance, and safety as carefully as possible, in line with regulated health system standards.”
Brian T. Horowitz began covering health IT news in 2010 and covered the entire technology beat in 1996.
