EY Canada has published a report on artificial intelligence adoption in Canada's public sector, finding that most organisations remain in the early stages despite strong interest from leaders.
The findings highlight a gap between ambition and deployment across government bodies. While 63% of public-sector leaders believe AI could significantly improve services, only 26% have deployed any AI tools.
Most organisations are still far from broad implementation. About 66% remain in planning or early development, and only 20% to 25% of government AI proof-of-concepts progress to scaled use.
As a result, many projects remain stuck at the pilot stage, a pattern the report describes as "PoC purgatory". EY Canada warns that stalled progress is creating risks for service delivery, operating costs and public trust at a time of rising demand for public services.
Adoption gap
The main obstacles are organisational rather than technical. Legacy systems, fragmented data, limited internal skills, and stricter requirements around privacy, transparency and accountability are identified as the main barriers to wider use.
These issues are especially relevant in government, where public bodies must show that new systems can operate securely, explain decisions and meet a higher standard of oversight than many private-sector users face. That makes it harder to move from experimentation to services used at scale.
EY Canada also examined agentic AI, a newer category of systems designed to act with greater autonomy. It found that only 10% of Canada's public-sector organisations have begun deploying agentic AI in any form, while 37% have no plans to use it at all.
The figures suggest that even as AI draws intense attention across the economy, more advanced use cases remain on the margins of public-sector adoption. In practice, many departments and agencies are still focused on assessing where conventional AI tools fit into existing service models.
Five steps
To address the slowdown, the report sets out a five-step framework for governments seeking to move beyond experimentation. First, define public outcomes clearly and choose use cases based on service problems rather than starting with the technology itself.
Second, build the foundations needed for broader deployment. That includes infrastructure, data governance and ethical frameworks that support secure and responsible implementation.
Third, design pilots with clear success measures and a path to wider rollout, rather than treating them as isolated experiments.
Fourth, prepare the workforce by engaging staff, developing skills and adjusting operating models so organisations can adopt and manage AI systems in day-to-day work.
Fifth, measure outcomes and share lessons across organisations to improve results and avoid repeating mistakes.
Public sector pressures
The report comes as governments face pressure to improve responsiveness while managing constrained budgets and ageing technology estates. AI is often presented as one route to better service delivery, but the findings suggest implementation remains slow and uneven across the sector.
That matters because the public sector relies on large, complex systems that are difficult to change quickly. Even where leaders support AI in principle, scaling it can require upgrades to data architecture, procurement changes, new governance structures and staff training before citizens see any impact.
The report also suggests that confidence in AI's potential has not yet translated into operational change. Leaders may see value in the technology, but adoption appears to stall when organisations confront the practical work of integrating tools into existing systems and proving they can meet public expectations.
In that respect, the findings portray a sector that is not resistant to AI so much as constrained by the structure of government itself. The central issue is less whether public bodies want to use AI than whether they can create the conditions needed to move projects from trial to everyday use.
With only around a quarter of organisations having deployed AI at all, and most proof-of-concepts failing to scale, the report presents a public sector still at the beginning of adoption rather than close to transformation.