Scaling Up: Commercializing Health AI in Canada

The following excerpts from the Fourth meeting of the C.D. Howe Institute’s Health Sector Economic Growth and Resilience Working Group

Meeting Overview

The C.D. Howe Institute’s Health Sector Economic Growth and Resilience Working Group met for its fourth session on June 23, 2026, to examine the adoption, oversight, and commercialization of artificial intelligence (AI) in Canada’s healthcare system. The meeting featured presentations from Shelley Spence, Auditor General of Ontario; Krista Balenko, Canada Health Infoway; and Gavin Tong, Accenture, followed by an open discussion among the group.

The panel examined a specific and rapidly scaling use case – AI-generated clinical documentation, or “AI scribes” – alongside the broader international policy environment that will determine whether this early point solution matures into a durable Canadian AI commercialization pathway. The meeting took place on the same day as a federal-provincial announcement extending investment in shared clinical data infrastructure to additional provinces,¹ which several participants cited as further evidence of growing momentum for AI in Canadian healthcare. They also acknowledged that public trust has not kept pace with adoption.

Participants broadly agreed that AI scribes have moved from pilot to meaningful scale faster, and with a more positive clinician reception than most had expected. At the same time, an independent audit presented at the meeting identified accuracy and governance gaps in the initial procurement of these tools, highlighting that oversight has not always kept pace with adoption. The discussion that followed centred on how Canada can safely accelerate the broader commercialization of healthcare AI, given a policy and procurement environment that most participants agreed remains fragmented relative to international peers.

Auditing Ontario’s First Wave of AI Adoption

Shelley Spence, Auditor General of Ontario

Shelley Spence presented findings from a recent audit of AI use across the Ontario Public Service (OPS), undertaken following the Ontario government’s 2024 AI strategy. The audit covered AI governance and oversight, the responsible and secure use of AI tools, and the procurement and evaluation of vendor-of-record AI systems. It included a specific deep dive into AI scribe systems used in healthcare and a separate facial-recognition-based document verification service.

At an OPS-wide level, the audit found that staff could access unsanctioned and unsecured AI tools from government-issued devices: approved AI tools were used at a rate of roughly 6 percent, against a 94 percent usage rate for non-approved generative AI websites, and only about 3 percent of staff had completed AI training as of August 2025.²

In the case of AI scribes, Supply Ontario procured the tools through a vendor-of-record model and made them available to physicians, nurse practitioners, therapists, and other clinicians.³ The audit found that security, accuracy, and bias-related criteria carried limited weight in vendor evaluation and that the process relied heavily on vendors’ own claims rather than independent verification, with no requirement for live product demonstrations.

Subsequent testing by clinical and technical staff from Ontario Health and the provincial Ministry of Health found that all 20 evaluated vendors exhibited at least one identified issue. More than half included incorrect medical information (including misidentified medications), 45 percent exhibited instances of “hallucinations” (fabricated content), and 30 percent had incomplete documentation or omissions. While these results are potentially concerning, participants noted that AI tools and their underlying algorithms are evolving rapidly and that the findings apply to tools available and in use through summer 2025. This rapid evolution also means that data security, output accuracy, and potential biases can change over time, requiring regular evaluation to ensure appropriate public sector use.⁴

The Auditor’s report recommendations focus on strengthening procurement discipline and improving policies and protocols for AI use within the public service, rather than curbing adoption. They include giving greater weight to security, privacy, and bias criteria in vendor scoring; requiring independent assessment of vendor security documentation rather than relying on vendor attestations; mandating live product demonstrations; requiring formal bias testing; and requiring clinicians to confirm they have reviewed AI-generated notes before those notes are finalized in the patient record. The Auditor General’s office follows up on recommendations after two and five years to track progress, and the government stated that they had already acted on some recommendations from this audit.

Scaling AI Scribes Nationally

Krista Balenko, Canada Health Infoway

Krista Balenko described a national initiative intended to provide up to 10,000 primary care clinicians with a 12-month funded AI scribe licence drawn from a list of pre-qualified vendors, targeting reach to roughly 20 percent of the primary care market. The program was built collaboratively at a national level, drawing on Ontario’s early vendor-of-record work as well as parallel efforts already underway in British Columbia and Nova Scotia. A national external advisory group set core clinical, business, privacy, and security requirements and participated in vendor testing and live demonstrations, an approach designed to avoid duplicating the same procurement and evaluation work in every jurisdiction. Registration was straightforward, directing all clinicians to a common landing page where they could select their region. From there, most jurisdictions opened access to any pre-qualified vendor, while Nova Scotia opted to deploy a single vendor province-wide. All regions were supported by common adoption resources such as readiness checklists, privacy guidance, and templates.

Demand exceeded expectations. More than 6,000 providers registered within two days of launch and more than 10,000 within two weeks, prompting several jurisdictions to close registration once notional allocations were reached. Balenko reported that the program has since supported more than 11.5 million patient encounters, at an average of roughly 1.2 million encounters per month – a volume that a large comparable US health system reportedly took some 15 months to reach with a similar enrolled user base.

An independent evaluation found strongly positive results across most direct measures: roughly 90 percent of clinicians rated the tool as valuable to their practice, about 80 percent reported a significant reduction in cognitive load, close to 70 percent reported reduced administrative burden, and about 80 percent reported feeling more engaged with patients (Centre for Digital Health Evaluation 2026). Roughly three-quarters of clinicians reported time savings. Of that group, about half redirected the time to completing other administrative or clinical tasks and roughly a fifth used it to see and/or take on more patients. Balenko cautioned that benefits are not uniform: the specific value delivered depends on the clinician, the appointment type, and the patient context.

Looking ahead, Balenko noted that clinicians increasingly view AI scribes as an entry point into a broader ecosystem of AI-enabled tools, including pre- and post-visit summaries, automated form completion and billing support, and clinical co-pilot functionality, several of which vendors are already layering onto existing products. In her assessment, the main policy question is how quickly AI scribes can be scaled responsibly – and under what governance and data-access arrangements.

Canada’s Commercialization Gap: An International Comparison

Gavin Tong, Accenture

Gavin Tong focused on the international economic policy environment shaping AI use cases in healthcare, particularly “front office” applications such as those involving personal health information and directly touching patient care, as distinct from valuable but less contentious “back office” applications. He described a broadly common two-part policy challenge facing every jurisdiction. The first is regulatory friction, including medical device certification, AI-specific regulatory regimes, and data privacy requirements. The second is creating commercialization pathways that convert regulatory approval into a viable, reimbursable route to market – historically a long and uncertain process for novel health technologies.

Tong described three distinct approaches that have been deployed across different jurisdictions. England has established a regulatory sandbox called the “AI Airlock,” allowing regulators and companies to jointly iterate on safe deployment controls for continuously learning AI medical-device products. Qualifying products move into a provisional clinical use and real-world monitoring phase before entering a national AI procurement framework, with current use cases concentrated in areas such as imaging triage, cardiac diagnostics, and dermatology (Medicines and Healthcare products Regulatory Agency 2026). Germany and France, by contrast, have focused on giving developers a clear route from regulatory approval to a “digital therapeutic” pathway that can be prescribed, with automatic access to national insurance coverage, applied particularly to mental health applications and combined patient-facing device and app products. Singapore offers a highly integrated model. It has placed AI at the centre of national economic policy since 2023, using a live national clinical dataset and sustained biotech investment to position itself as a launch pad from which private companies can scale innovations across the wider Asia-Pacific market.

Against this backdrop, Tong argued that Canada holds many of the underlying ingredients for success: strong AI research, rich longitudinal health data, a receptive and well-educated workforce, and an early proof point in AI scribes. Canada’s challenge is that it lacks the commercialization architecture to leverage these foundational assets into rapid commercialization, adoption, and scale. Even if Canada achieves a more unified national approach to regulation, he suggested, commercialization is likely to remain fragmented across 13 separate provincial and territorial procurement processes. Healthcare also has historically had a risk-averse purchasing culture and a reluctance to be a first customer for novel technology, meaning many companies launch their products and capture value abroad before entering the Canadian market.

The presentation also raised concerns about AI-driven workforce displacement across both blue- and white-collar occupations and the well-documented health consequences of concentrated unemployment, particularly among younger workers, as a longer-term risk that merits attention alongside the immediate commercialization of AI tools. Future adoption patterns and applications of new AI tools remain uncertain. Given workforce and access challenges in healthcare, improving the efficiency of healthcare delivery will become increasingly important, alongside monitoring AI’s broader and evolving economic and health impacts.

Policy Discussion

Opening the discussion, the moderator summarized the tension raised across the three presentations: a point solution – AI scribes – that has already achieved wide and enthusiastic clinician uptake despite documented imperfections in how it was first procured, set against a more uncertain outlook on Canada’s broader capacity to commercialize healthcare AI. The group was asked how to create room to accelerate commercialization while maintaining accuracy, security, and bias safeguards, and whether Canada’s best path is to learn from international leaders, even at the cost of being a “second or third mover,” or to pursue a made-in-Canada sandbox and national procurement model.

Several participants argued that Canada should embed AI directly in its own regulatory processes, rather than only regulating AI as a product, pointing to international examples where AI-assisted application review has replaced slower, correspondence-based bureaucratic processes to shorten approval timelines without weakening oversight. This was linked to a recurring and largely undisputed observation: participants repeatedly identified the fragmentation of healthcare procurement and regulatory authority across 13 provincial and territorial jurisdictions as the primary structural obstacle to scaling AI commercialization in Canada, in contrast to the single national markets of peer jurisdictions such as Germany and France. While this structure is constitutional, provinces can still cooperate to make the Canadian market more attractive by harmonizing regulatory and procurement policies and processes.

One participant drew a direct comparison to Canada’s national shipbuilding strategy, in which a single centralized federal buyer aligns procurement with domestic economic development goals, and asked how a similar alignment mechanism might be built across health technology purchasing – including a “buy Canada” dimension – given the absence of an equivalent central decision-maker in healthcare. Another participant cautioned against treating this purely as a procurement design problem, noting that jurisdictional fragmentation reflects broader structural features of Canadian health federalism that a procurement fix alone would not resolve. The group did not fully reconcile these two framings.

A further international example was offered as a model for policy learning rather than pure imitation: Germany and France’s collegial “leapfrogging” on digital therapeutics reimbursement frameworks. Germany’s early rollout identified a gap in clinician and patient training that France subsequently addressed in its own framework, with the two countries openly exchanging implementation lessons.

On the regulatory classification of clinical AI tools, one participant noted that AI scribes generally sit outside Health Canada’s medical device framework, unlike more clinically substantive diagnostic or decision-support AI applications, though guidance specific to generative and higher-risk AI tools continues to develop. The central challenge, in this view, is calibrating oversight to a tool’s genuine clinical risk and value so that governance does not simply add to clinician burden. The participant also suggested exploring national procurement principles that could sit alongside continued provincial responsibility for funding and administration, rather than requiring a full redesign of health system financing.

Echoing this, another participant – drawing on work developing a “patient charter” through a co-design process – suggested that Canada’s governance toolkit of legislation, policy, and programmatic guidance is not yet clearly organized to distinguish between areas where absolute certainty is required and those where flexibility should be preserved while the technology and public confidence in it continue to evolve.

The discussion closed with a participant cautioning that while AI tools, including scribes, offer clear benefits, public and clinician trust remains uneven. The participant cited an example in which a physician declined to see a patient who had asked that an AI scribe not be used, illustrating a live tension around patient choice and informed consent that adoption figures alone do not capture. These tensions must be balanced with reported clinician-side gains, including several hours of weekly time savings and more face-to-face time with patients. Participants broadly agreed that healthcare AI remains a moving target for regulatory policy and procurement-stage product validation and, at least for now, requires human validation and verification.

Policy Conclusions

There was clear consensus that AI scribes have demonstrated clinician demand and delivered measurable value in clinical practice. Participants generally viewed the risk of under-investing in healthcare AI, given capacity and access challenges, as greater than the risks of broader adoption, provided accuracy, bias, and security safeguards keep pace with deployment. Participants likewise agreed that human review remains a necessary safeguard against the documented error rates in current AI scribe products, and that Canada’s fragmented procurement and regulatory landscape across 13 provincial and territorial jurisdictions is the principal structural barrier standing between early wins in AI adoption and a durable Canadian AI commercialization advantage. Canada Health Infoway’s recent AI Scribe Program demonstrated that a successful and scalable national procurement and coordinated deployment process is possible, offering a potential model for other products.

Genuine uncertainty remains in several areas. Participants could not say with confidence how much AI use is already occurring outside sanctioned or governed channels, noting that this kind of activity is inherently difficult to observe or quantify. There was similar uncertainty about whether Canada can build a coordinated national approach to regulation and procurement quickly enough to keep pace with faster-moving peer jurisdictions, or whether a province-by-province default will persist. Multiple participants noted that the pace, scale, and distributional impact of AI-driven workforce displacement will need to be monitored, along with associated effects on population welfare and healthcare needs.

There was some debate about Canada’s ideal adoption and regulatory strategy. Some participants favoured a deliberate “fast follower” posture, learning from international leaders such as England, Germany, France, and Singapore. Others argued that this posture risks entrenching the very lag participants identified as Canada’s central weakness and favoured a more assertive domestic approach combining a Canadian regulatory sandbox with coordinated national procurement. Participants also differed on whether persistent 13-jurisdiction fragmentation could be addressed through a centralized buying model akin to national defence procurement, or as a symptom of deeper structural features of Canadian health federalism that procurement reform alone cannot resolve. Overall, participants agreed that regulatory modernization is a necessary companion to continued AI scribe scale-up, and a precondition for translating Canada’s research and data advantages into a genuine commercialization pathway.

Participants of the C.D. Howe Institute Health Sector Economic Growth and Resilience Working Group

Individuals participate in their personal capacities, and the views collectively expressed do not represent those of any individual, institution, or client.

Co-Chairs:

Helen Angus, CEO, AMS Healthcare

Tony DiEmanuele, President and CEO, Mohawk Medbuy Corp. (session chaired by Peter Longo, COO, Mohawk Medbuy Corp., in Mr. DiEmanuele’s absence).

Presenters:

Shelley Spence, Auditor General, Office of the Auditor General of Ontario

Krista Balenko, Vice President, Enablement and Operations, Canada Health Infoway

Gavin Tong, Canadian Health Industry Data and AI Lead, Accenture

Participants:

• Neala Barton, VP, Communications and Client Experience, Canadian Institute for Health Information
• Brigid Buckingham, Director of Government Affairs and Policy, GE Healthcare
• Barbara Carnegie, Vice President, Healthcare and Group Benefits, Canadian Life and Health Insurance Association
• Nathan Clark, Director, Government Relations, Manulife
• Donna Duncan, CEO, Ontario Long Term Care Association
• Jonathan Feairs, Head, Government Affairs and Enterprise Policy, AstraZeneca Canada
• Cathy Hecimovich, CEO, Ontario Retirement Communities Association
• Jim Keon, President, Canadian Generic Pharmaceutical Association
• Geneviève Lavertu, Government Affairs and Policy, Johnson & Johnson Canada
• Richard Lech, Senior Manager, Market Access, Edwards Lifesciences
• Leslie Madden, Head of Government Affairs and Policy, Moderna
• Chad Mitchell, Senior Association, Global Public Affairs
• Janice Mundell, Chief Marketing Officer and SVP of Business Intelligence and PMO, Mohawk Medbuy Corporation
• Christian Ouellet, VP Corporate Affairs, Sandoz Canada
• Paul Petrelli, Vice President and General Manager, Gilead Sciences Canada
• Joanne Pitkin, Vice President of Government and Public Affairs, GreenShield Canada
• Ramesh Ranjan, Lead, Official Engagement, Canadian Blood Services
• Nissa Shariff, Partner, Health, Government and Public Sector, PricewaterCoopers LLP
• Lori Shortill, Sales Excellence Leader, Philips
• Dan Singh, Director, Strategy and Innovation, Canada’s Drug Agency (on behalf of Suzanne McGurn, CEO)
• Michelle Theroux, Partner, Life Sciences & Healthcare, Deloitte

References

Auditor General of Ontario. 2026. Performance Audit: Use of Artificial Intelligence in the Ontario Government. https://www.auditor.on.ca/en/content/specialreports/specialreports/en26/2026_AI_EN.pdf.

Centre for Digital Health Evaluation, Women’s College Hospital Institute for Health System Solutions and Virtual Care. 2026. “Evaluating Artificial Intelligence (AI) Scribes to Reduce Administrative Burden and Enhance Data Interoperability in Canadian Primary Care.” https://insights.infoway-inforoute.ca/docs/component/edocman/497-results-report-evaluating-artificial-intelligence-ai-scribes-to-reduce-administrative-burden-and-enhance-data-interoperability-in-canadian-primary-care?Itemid=0.

Medicines and Healthcare products Regulatory Agency. 2026. “AI Airlock: The Regulatory Sandbox for AIaMD.” https://www.gov.uk/government/collections/ai-airlock-the-regulatory-sandbox-for-aiamd.