Ottawa fast-tracks AI in public service to cut costs, improve access, and anchor digital sovereignty

Ottawa's AI Push: Practical Steps for Modernizing Public Service

Ottawa is moving to accelerate the use of artificial intelligence across the public service. The goal is straightforward: modernize operations, cut costs, and make it easier for Canadians and public servants to find accurate information fast.

Louis Tetu, chair at Coveo, says the clearest near-term wins are AI-search and generative tools that help people self-serve and help employees work smarter. Below is a practical playbook for government teams to turn that intent into measurable outcomes.

Key Takeaways

• AI-search and generative tools can improve self-service, agent assist, and internal knowledge access.
• There is a meaningful cost-savings opportunity and a chance to give time back to citizens.
• Digital sovereignty matters: keep data in Canada and prioritize Canadian-built infrastructure where it makes sense.
• Government can act as a catalyst for wider AI adoption across the Canadian economy.
• Retaining and industrializing Canadian AI talent is essential for productivity and competitiveness.

Why This Matters for Public Servants

Most service friction is information friction. Citizens can't find answers. Agents dig through multiple systems. Policies are buried in PDFs. AI-search and retrieval-augmented generation can unify content and return precise, source-cited answers instantly.

This isn't about flashy chatbots. It's about reducing call volumes, improving first-contact resolution, and shortening time-to-answer for both citizens and employees.

Where to Start: Low-Risk, High-Impact Pilots

• Citizen self-service: Layer AI-search over existing web content and service portals. Focus on top 50 high-demand tasks (benefits, visas, tax, permits). Return answers with citations to official sources.
• Agent assist for contact centres and casework: Summaries, next-best steps, and policy snippets with links back to the authoritative page or clause.
• Internal knowledge search: Unite GCdocs, SharePoint, service desks, and policy repositories to reduce swivel-chair time and duplicate tickets.
• ATIP triage and drafting assist: Classify requests, suggest drafts from approved templates, and flag sensitive content for review.
• HR and IT support: Virtual assistants for common requests (passwords, onboarding, equipment, leave) with clear escalation paths.

Guardrails by Design: Privacy, Security, Accuracy

• Data residency: Keep sensitive data and model inferences in Canada. Require encryption in transit and at rest.
• No uncontrolled data sharing: Prevent uploads of personal or protected information to public models. Use private, monitored endpoints.
• Model evaluation: Measure precision/recall, citation correctness, and hallucination rates per use case. Set thresholds before go-live.
• Human-in-the-loop: Require review for sensitive outputs (eligibility decisions, compliance guidance) and log decisions for audit.
• Accessibility and inclusion: Enforce WCAG 2.1 AA and test outputs across official languages.
• Follow policy: Use the Government of Canada's Directive on Automated Decision-Making and complete an Algorithmic Impact Assessment where needed. See official guidance here.

Digital Sovereignty: Build on Canadian Infrastructure

There's strategic value in having Canadian-built AI infrastructure and keeping sensitive workloads inside the country. Tetu highlights this as a core priority: efficiency for government and a catalyst for the broader economy.

In practice, that means Canadian data centres, clear data residency commitments, and vendors that support open standards. Avoid tight coupling to a single model. Design for portability so you can swap models without rewriting everything.

Talent: Retain, Develop, Industrialize

Canada trains exceptional AI talent but often loses it to foreign employers. The fix is straightforward: create compelling public-sector roles, meaningful missions, and clear career paths that blend policy expertise with AI delivery.

• Stand up small AI product teams (product, data, engineering, policy, security) inside priority programs.
• Fund ongoing training and certifications for analysts, PMs, and technical staff. Track skills developed per quarter.
• Build communities of practice across departments to share patterns, benchmarks, and reusable components.

If you need structured upskilling aligned to job roles, explore focused learning paths by role here.

Procurement That Moves

• Outcome-based RFPs: Specify target metrics (deflection rate, AHT reduction, satisfaction) rather than features.
• Start small, scale fast: 90-day pilots with clear exit and expansion clauses.
• Security and compliance baked in: Privacy Impact Assessment, TRA, data residency, records management, accessibility, bilingual outputs.
• Interop first: APIs, SSO, logging, and connectors to existing systems. No black boxes.

Measure What Matters

• Citizen self-service: Task completion rate, containment/deflection, time-to-answer, and satisfaction scores.
• Agent assist: Average handle time, first-contact resolution, case backlog, and training time for new staff.
• Internal knowledge: Search success rate, duplicate tickets avoided, time saved per query.
• Quality: Citation accuracy, factual error rate, and escalation frequency.

Set baselines before launch. Review weekly during pilots. Publish results to maintain momentum and trust.

90-Day Delivery Plan

• Days 0-30: Pick two use cases. Confirm data sources and guardrails. Stand up secure environment and connectors. Draft content standards.
• Days 31-60: Build prompts, retrieval pipelines, and evaluator tests. Train staff. Start red-teaming and bilingual testing.
• Days 61-90: Pilot with a defined audience. Track metrics daily. Fix gaps. Prepare a scaling plan with clear funding and support.

What Louis Tetu Emphasized

• There's immediate value in helping Canadians self-serve and helping public servants access better information.
• Cost savings could be significant, and the bigger win is time returned to citizens and staff.
• Government leadership can spur adoption across the economy, especially with a sovereign AI infrastructure.
• Stop subsidizing foreign hiring of Canadian talent. Grow domestic capacity and keep it inside Canadian institutions and businesses.

Common Pitfalls to Avoid

• Over-indexing on chatbots without strong retrieval and content quality.
• Deploying models without evaluation, guardrails, or auditability.
• Lock-in to a single vendor or model with no portability plan.
• Ignoring bilingual performance and accessibility from day one.

Next Steps for Departments

• Appoint an AI product owner and assemble a small cross-functional team.
• Select two pilot use cases and define success metrics upfront.
• Complete required assessments and confirm data residency and security controls.
• Launch a 90-day pilot, measure weekly, and scale based on evidence.

Government doesn't need more hype. It needs working systems that save time, reduce costs, and improve service. Start with AI-search, build strong guardrails, and grow the wins from there.