evidoria

← Back to browse

Good practice Imported

Ontario's Microsoft 365 Copilot Rollout — Productivity Gains Undercut by Shadow AI and Skipped Training

Canada · Toronto · See the Canada profile · See the Toronto profile

Evidence: Descriptive / self-reported Top 75% 47/100 · Ask Evidence Copilot about this practice

Weekly Copilot Chat use by 15,000+ Ontario public servants saved nearly 3 hours/week each, but the province's Auditor General found the sanctioned tool handled just 6% of staff AI use, versus 94% on unsanctioned sites, with only 3% trained.

15,000+ employees/week
Weekly active users of sanctioned Copilot Chat (late 2025)
55,000 employees
Total civil servants addressed by rollout
~3 hours/week
Self-reported average weekly time saved per employee
120,000+
Portal pageviews (by late 2025)
6 %
Share of staff AI usage on sanctioned platform (April–August 2025)
94 %
Share of staff AI usage on unsanctioned external tools (April–August 2025)
3 % (1,800 of 55,000)
Staff who completed mandatory Responsible Use of AI training (by August 2025)
Ontario's Microsoft 365 Copilot Rollout — Productivity Gains Undercut by Shadow AI and Skipped Training

Details

Maturity
Scaling
Promoter
Government of Ontario / Ontario Public Service
Period
2024 (pilot) – ongoing; Auditor General audit report May 2026
Keywords
generative AI, Microsoft 365 Copilot, public-sector productivity, AI governance, shadow IT

Context

Ontario deployed Microsoft 365 Copilot Chat via an 'InsideOPS' portal to approximately 55,000 civil servants, reporting the highest Copilot usage among Canadian governments. By late 2025, over 15,000 weekly users generated 120,000+ pageviews, and Minister Stephen Crawford reported average weekly time savings of almost three hours per employee, targeting a 20% productivity uplift.

Results

An Ontario Auditor General special report (May 2026), examining AI usage between April and August 2025, found that the sanctioned Copilot Chat platform accounted for only about 6% of total staff AI usage, while unsanctioned external AI tools accounted for 94%. Only about 1,800 of 55,000 staff (roughly 3%) had completed mandatory Responsible Use of AI training by August 2025.

Conclusions

The case is mixed: productivity-time-savings claims for the sanctioned tool are self-reported by the Minister, but the scale of shadow-AI use and the training shortfall were independently confirmed by the Auditor General, revealing a governance gap alongside the productivity claims.

Implementation
Indicative cost
High (€500k–€5M) — Microsoft 365 Copilot Chat licensing at tens-of-thousands-of-seats scale, likely representing a substantial recurring cost, though no figures were published.
Time to results
Medium (1–3 years) — Pilot from 2024, expanding through 2025; independent Auditor General audit covering April–August 2025 published May 2026.
Staffing & skills
Ontario Public Service digital/IT team operating the InsideOPS portal, Office of the Auditor General of Ontario (independent audit function)

Conditions for success

  • Sanctioned, secure platform (Microsoft 365 Copilot Chat) made available through a dedicated internal portal
  • Mandatory Responsible Use of AI training programme (though completion lagged)

Common failure modes

  • Only ~6% of actual staff AI usage ran through the sanctioned tool; 94% used unsanctioned external sites, creating documented data-governance risk
  • Only ~3% of staff completed mandatory AI training by the audit period
  • Productivity-savings figures are self-reported by government, not independently audited

Where it fits

Governance type
sub-national (provincial) government
Scale
large provincial public service (~55,000 staff)
Income level
high income (Canada)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful