Health systems face pressure from labour shortages, administrative complexity, ageing populations and changing regulation. AI adoption is increasing, but measurable value remains uneven. A Q4 2025 McKinsey US Gen AI Healthcare Survey found that 50 percent of healthcare leaders said their organisations had implemented generative AI, while only 45 percent of those organisations had quantified return on investment. Current use remains concentrated in areas such as coding, appeals and ambient listening. The operational challenge is to move from isolated pilots to redesigned workflows, clearer governance and regular performance review.
From AI Adoption to Organisational Priorities
AI implementation needs a clear leadership role, because fragmented adoption can leave organisations with visible activity but limited measurable effect. CEOs set the level of ambition, connect AI to organisational priorities and align capital allocation, governance and talent around a shared direction. That link also helps prevent AI work from becoming a separate digital programme. Without sustained sponsorship, individual initiatives can remain narrow, incremental and disconnected from wider operational needs.
Health systems also need to choose where AI can support domain-level change. A tool that helps draft appeal letters may reduce work in one task, but it offers less value than coordinated AI agents supporting sequential administrative workflows across claims, payment processing and reconciliation. The same distinction applies across other operational areas where handoffs and manual steps limit the effect of isolated tools.
Selecting one or two high-value domains gives AI work a clearer operational focus. Administrative workflows and supply chain processes are potential starting points because they involve sequential tasks, multiple handoffs and limited clinical risk. More complex areas, including care access and workforce management, can follow as organisational capability develops. This sequencing keeps early implementation close to practical operational needs while leaving room for gradual expansion into more demanding care-related domains.
Redesigning Workflows Around Defined Guardrails
AI creates greater value when health systems redesign work around required outcomes rather than insert technology into legacy processes. End-to-end transformation means redefining what work is needed, then rebuilding roles, processes and technology around that work. This approach shifts attention from tool deployment to how patients, caregivers and the organisation experience the redesigned workflow.
Technical foundations matter, but they do not need to be complete before any operational change begins. Waiting for perfect data, infrastructure and tooling can delay progress. Reusable data and infrastructure can develop alongside priority domains, with investment guided by actual business needs. Building foundations in that sequence can reduce the risk of creating technical assets that remain detached from daily work.
Must Read: Reliable AI Document Processing Needs Operational Proof
Governance and risk management require earlier clarity. Health systems need a simple set of guardrails and processes so that everyone involved understands the organisation’s principles and the boundaries for AI use. Governance should provide a clear philosophy and an evaluation process for applications, rather than rely only on rigid rules. A more flexible approach can help organisations apply principles consistently as AI changes.
Cross-functional product development pods can support this redesign. These groups bring together designers, end users, subject matter experts and technologists. Business leaders can guide pods because they understand outcomes, operational issues and process nuances, while product owners remain accountable for prioritisation, resources, process design and value measurement.
Measuring Progress and Building AI Fluency
Each domain needs near-term and long-term performance metrics tied to value creation. Executive teams, including the CEO, should review those metrics quarterly. In administrative workflows, value drivers can include improved documentation accuracy, lower collection-related costs, yield improvement and faster cash flow. Long-term indicators can measure outcomes linked to those drivers, while short-term indicators can track the actions needed to reach them.
Care and case management require different measures. Value drivers can include fewer hospital readmissions and fewer unnecessary emergency room visits. Near-term indicators may include completed care transition assessments, medication reconciliation within 48 hours of discharge, follow-up visits scheduled within seven days and acute exacerbations predicted at least two weeks in advance. These measures connect AI-enabled workflow change to observable operational and care-management steps.
Underperformance needs prompt attention. Once leadership has set expectations for pace and progress, pod owners can remove barriers, stop low-yield work or pivot when indicators are not being met. This approach supports learning, experimentation and redeployment of talent to more productive opportunities.
AI fluency also needs to extend beyond specialist teams. Leaders, middle managers and frontline workers need enough understanding to identify opportunities and collaborate with technologists. Training, space for experimentation and financial or non-financial incentives can support adoption, especially early in implementation when established processes can reassert themselves.
AI impact in health systems depends on disciplined management rather than technology adoption alone. The core task is to connect AI to priority domains, redesign workflows, establish guardrails, mobilise cross-functional teams and track progress with relevant measures. Isolated pilots can show activity without producing measurable organisational change. A more operational approach gives health systems a clearer route from implementation to value, while keeping attention on patients, caregivers, workforce needs, financial sustainability and practical execution rather than isolated experimentation across priority domains.
Source: McKinsey&Co
Image Credit: iStock
