error
Subscriptions are not available for this site while you are logged into your current account.
close
Skip to main content

      Canadian health organizations have no shortage of interest in artificial intelligence. Pilots are important: they create room to learn, test clinical value, understand workflow implications and uncover risks before wider deployment. The problem is what happens when a pilot works.

      A successful pilot answers an important question: can this use case create value in a defined setting? Scaling has to answer a harder one: can the organization reproduce that value across teams, sites and patient populations, while supporting clinicians, managing risk and sustaining performance over time?

      That transition is where many AI initiatives become difficult. The capabilities required to prove an AI use case are not the same as the capabilities required to operate it as a reliable clinical service.  

      The hard part starts after the pilot

      The implementation gap is visible in the evidence. A 2025 systematic review of 1,263 intensive-care AI studies found that 74% remained in early development or validation stages and only 2% progressed to clinical integration.1 That is not a Canadian hospital benchmark, but it illustrates how much easier it is to develop and validate AI than to embed it into routine care.

      Canadian guidance points in the same direction. Health Canada’s 2026 guidance for machine-learning-enabled medical devices treats AI as a lifecycle challenge: clinical validation, workflow fit, risk management, human factors, post-market monitoring and the possibility of performance degradation or data shifts after deployment all matter.2

      The implication for health-system leaders is practical. Go-live is not the finish line. It is the start of a new operating responsibility - one that needs sustained ownership, support and measurement.

      A successful pilot should be the beginning of the value conversation, not the end of it.

      Build a clinical AI translation capability

      Hospitals already know how to run projects, implement technology and analyze data. Many also already have the ingredients needed to scale AI: clinical informatics, quality and safety, digital, analytics, privacy, cybersecurity, operations and change expertise. The challenge is that these capabilities often sit in separate functions with different mandates, priorities and accountabilities. Scaling clinical AI requires them to work across those organizational seams rather than being reassembled for each pilot.

      That does not automatically mean creating another large centralized AI team. Some organizations may need new roles or dedicated capacity; others may need clearer cross-functional structures, decision rights and role accountabilities that connect capabilities already in place. The structural answer will vary by hospital, but the ownership model cannot remain implicit.

      Lydia Lee

      Global Co-Lead, KPMG Connected Enterprise for Health, KPMG International; Partner and National Leader, CIO Advisory Services and Digital Health

      Toronto

      KPMG Canada




      Five things to put in place before scaling a successful pilot

      • Assign end-to-end ownership

        Name one accountable owner for the use case from problem definition through deployment, adoption, value realization, ongoing support and eventual retirement. The role may sit with an existing clinical, operational or digital leader, but the accountability should be explicit: is this still improving the clinical problem it was intended to solve?

      • Design the workflow and adoption model for scale

        Design how the technology will work in routine care before broad rollout. Involve end users, clarify decision rights, train teams and manage behaviour change across clinical operations, informatics, education and transformation. AI that performs well but does not fit the way care is delivered will struggle to create sustained value.

      • Put production AI operations in place

        Ensure there is accountable engineering capacity to deploy, integrate, version, monitor and maintain models and AI services in production. Translational clinical AI literature distinguishes this work from data science itself and highlights the role of ML engineering and MLOps in moving tools into routine practice.3 For many hospitals, the question is less whether the expertise exists somewhere than whether it is available, funded and accountable for ongoing clinical AI operations.

      • Establish risk-based clinical assurance

        Bring clinical safety, privacy, cybersecurity, data governance, ethics and regulatory expertise together through a practical, proportionate pathway. Ontario's AI directive provides a useful principle: determine risk level and apply risk-based proportional controls.4 These functions may already exist, but they need a mechanism to make timely, coordinated decisions. Higher-risk clinical uses warrant deeper scrutiny; lower-risk uses should not face identical process burdens.

      • Define and continuously measure clinical value

        Define the clinical outcomes that justify scale, then keep measuring them after deployment: patient outcomes, preventable harm or errors, clinical workflow, adoption and clinician time. This requires an enduring link between analytics, quality, clinical leadership and operations, not a one-time pilot evaluation. Monitor across sites and populations rather than assuming a local result will reproduce itself automatically.



      Measure progress by clinical value, not AI activity

      The pressure to “keep up” with AI can make visible activity feel like progress. A growing list of pilots may reassure executives, clinical leaders or boards that an organization is moving. But pilot volume is a weak proxy for impact if promising use cases cannot make the transition into routine care.

      A stronger portfolio starts with a small number of health-system objectives. Is the organization trying to reduce preventable harm? Improve patient outcomes? Give clinicians time back for care? Reduce documentation burden? Those goals should shape which use cases are funded, how pilots are evaluated and which ones earn the organizational capacity required for scale.

      Documentation is a useful example because the value is tangible. In the Canadian Medical Association’s 2025 physician survey, physicians reported 10.4 hours per week of administrative work; among physicians using AI, 59% said it had reduced their administrative time.5 The bigger question is whether that time saving persists, is distributed fairly across users and ultimately creates more capacity for patient care.

      Governance should accelerate safe scale

      Clinical AI needs appropriate governance, and many organizations are still in the early stages of putting it in place. The practical issue is whether governance helps the organization make better, faster and more consistent decisions, or becomes another hand-off that slows implementation without materially reducing risk.

      A risk-based model is more useful: establish clear enterprise guardrails, standard evidence requirements and delegated pathways for lower-risk uses, while reserving deeper multidisciplinary review for applications that can materially affect clinical decisions or patient outcomes. The role of an AI governance committee should be to set the rules of the road, resolve higher-risk questions and oversee the portfolio, rather than supervise every implementation decision.

      If a health AI pilot proved valuable tomorrow, would your organization have the product ownership, workflow expertise, operating capability, risk-based assurance and measurement capacity to scale it?

      Scaling AI is an enterprise leadership responsibility

      AI scale cannot be delegated to the CIO, CMIO, chief digital officer or a clinical operations leader. Once an AI use case moves into routine care, it can affect clinical practice, workforce capacity, risk, capital and operating priorities. Those are enterprise choices. The executive team needs shared ownership of the outcomes being pursued, the capacity committed to scale and the trade-offs required to sustain adoption.

      Boards also have an important role, although not in approving individual use cases. They should expect management to be clear about why the organization is investing in AI, how value and patient risk are being measured, where accountability sits and how higher-risk clinical applications are being overseen. Successful scale should be visible in the organization's performance and risk dialogue, not treated as a technology program operating off to the side.

      The practical question for leaders is therefore broader than whether the organization can run an AI pilot. If a promising pilot demonstrated meaningful clinical value tomorrow, would the organization know who owns the decision to scale it, have the cross-functional capacity to operationalize it, and keep the executive team and board focused on whether that value is sustained?

      That may be the more useful test of AI readiness - and a better measure of progress than the number of pilots underway.

      Sources

      1. JAMA Network Open. Operationalization of Artificial Intelligence Applications in the Intensive Care Unit: A Systematic Review. 2025.

      2. Health Canada. Pre-market guidance for machine learning-enabled medical devices. 2026.

      3. npj Digital Medicine. An operational guide to translational clinical machine learning in academic medical centers. 2024.

      4. Government of Ontario. Responsible Use of Artificial Intelligence Directive. current page accessed 2026.

      5. Canadian Medical Association. Pressure and progress: Results of the CMA’s 2025 National Physician Health Survey. 2025.

      6. Vector Institute. Health AI Implementation Toolkit. accessed 2026.

      7. KPMG Canada. Using artificial intelligence to optimize care delivery systems. 2025.

      8. KPMG Canada. Beyond AI adoption: Turning Canada’s AI momentum into measurable returns. 2026.


      Insights

      AI is emerging as a powerful ally, enhancing how we plan and deliver care.

      Why investing in people, process, and governance is essential to turning AI adoption into sustained value.


      Connect with us

      KPMG. Make the Difference.

      We’re here to help your organization thrive.

      Office building