Inova’s AI pipeline moved from monthly reviews to enterprise-scale oversight
Two years ago, Inova Health System was reviewing about one artificial intelligence feature each month. Today, the Falls Church, Va.-based system evaluates more than a dozen AI proposals monthly and already has 92 AI features in production. The pace of change has forced the organization to build a governance model that can absorb rapid growth without slowing deployment.
The shift is not just a matter of volume. It is a structural change in how a health system decides what deserves to move forward, what should be monitored after launch and what should be shut down when it no longer performs as expected. Jon McManus, vice president and chief data and AI officer at Inova, said the system is preparing its governance infrastructure to oversee as many as 500 AI tools.
One sentence captures the operational pressure inside the organization: the intake process is no longer the bottleneck; the scaling problem is.
Governance now has to keep pace with a portfolio that keeps growing
Inova’s monthly review count has climbed from roughly three or four AI features a year ago to more than 12 today. Those proposals come from several directions at once: new capabilities from existing technology vendors, employee-generated ideas and collaborations with outside companies, including OpenAI and Anthropic. The result is a continuous flow of potential projects that must be screened for value and risk before they reach production.
The health system evaluates each proposed AI feature against four pillars: value, safety, reliability and impact. It also measures whether the idea fits organizational priorities such as growth, workforce well-being and cost resilience. Those criteria are designed to prevent enthusiasm for AI from outrunning operational discipline.
Inova has reached a point where demand can exceed implementation capacity. Rather than allowing that gap to create chaos, the system built an AI governance platform to handle reviews and oversight at scale. McManus said the goal is to avoid reverting to spreadsheets as the portfolio expands. The platform uses AI agents to help with market intelligence and the preparation of governance documents, but human reviewers remain part of the process.
Distributed implementation with centralized oversight
Inova does not run every AI project through a single centralized implementation team. Instead, it distributes execution across the organization while keeping governance and post-deployment monitoring centralized. That model allows more than 30 AI projects to move forward at the same time, according to McManus.
The structure matters because AI adoption in health systems often fails when every request is forced through one overburdened team. Inova’s approach separates the work of building and deploying tools from the work of deciding whether those tools deserve to exist in the first place. It also preserves a single set of standards for monitoring once a project goes live.
A dedicated AI product management team handles vendor oversight, administration, risk and governance. A separate AI solution design team helps departments convert early concepts into technical requirements before engineering teams get involved. In practical terms, that means a clinician or operational leader does not simply bring an idea to engineers and hope for the best. The project is shaped, categorized and routed before development begins.
The system also distinguishes between vendor-based products and internally built applications. Ideas that appear to fit an existing vendor solution move through product management and governance workflows. Internally developed projects receive extra design support so teams can identify the right platform and technical requirements before engineering starts.
EHR-centered projects follow one path, external tools another
For projects tied to changes inside the electronic health record, Inova favors Epic’s agent platform. Other projects may use OpenAI, Anthropic or additional technology partners, depending on the technical and operational need. The platform choice is not cosmetic. It reflects where the work will live, how it will be integrated and how much redesign it will require across the workflow.
This separation creates a more realistic deployment path for a portfolio that spans multiple use cases. Not every AI tool should be built or implemented the same way, and Inova’s governance process appears designed around that reality rather than around a single favored architecture.
The model also places a premium on translation. Ideas that originate in clinical departments or operational teams do not move directly into code. They are converted into technical requirements first, which reduces the chance that enthusiasm for a concept outruns its buildability.
AI literacy has become a workforce requirement
Technology alone does not solve the scale problem. McManus said AI literacy is the key differentiator for managing decision-making at scale. He framed the issue as an enterprise-wide responsibility rather than a task that can be delegated to the AI office alone.
Inova has built AI education programs for employees, clinical teams and organizational leaders. Its leadership curriculum teaches executives how to identify problems that are appropriate for AI and how multiple AI applications can work together to address larger operational challenges. The emphasis is on using existing investment and resource-allocation processes instead of creating an entirely separate bureaucracy for every new AI initiative.
That approach recognizes a basic operational truth: AI becomes useful only when leaders understand where it fits, where it does not and how one tool may interact with another inside a busy health system. Without that literacy, governance can become either too rigid or too permissive. Neither outcome helps clinicians, staff or patients.
Quarterly monitoring can end with decommissioning, not just deployment
Inova’s oversight does not stop once an AI feature goes live. The health system conducts quarterly safety and reliability reviews of production AI tools, combining feedback from operational owners with quantitative performance measures. Those reviews are not ceremonial. They are meant to answer whether a tool is still meeting expectations in real-world use.
The most recent quarterly review identified three AI features whose performance scores had fallen enough to require additional committee scrutiny. Inova developed decommissioning plans for all three.
That decision is one of the clearest signs of maturity in the program. Many health systems focus on activation, pilot launches and expansion. Inova is also building a process for turning tools off when they no longer perform adequately. In a domain where software can affect workflows, decision support and operational reliability, decommissioning is as important as deployment.
McManus described that step as a major milestone: organizations are not just turning AI on; they are purposefully turning it off when performance standards are no longer met. The statement reflects a governance model that treats AI as a managed clinical and operational asset, not as a permanent fixture once installed.
Other health systems will need more than enthusiasm to keep up
As Inova prepares for a portfolio that may grow to 500 AI tools, McManus said other health systems should strengthen workforce capabilities and reinforce existing decision-making processes before creating entirely new structures around AI. The message is direct. Institutions do not need a parallel universe of committees to manage artificial intelligence; they need disciplined governance, clear roles and staff who can evaluate tools intelligently.
The article leaves one important operational question open: how many health systems can build this level of oversight before the volume of AI proposals outpaces their internal capacity. Inova’s answer is to centralize governance, distribute implementation and monitor performance continuously. Whether other systems can match that combination will depend on staffing, leadership alignment and how quickly AI moves from novelty to routine infrastructure.