AI Consulting for the Federal Government of Canada

When federal teams call us

This page is for branches that have a mandate to use AI and a short window to show a plan. Typical situations:

• A program area wants AI in case processing. The team needs to know early whether the Directive on Automated Decision-Making applies and which impact level to expect.
• Staff already use generative AI for drafting. The branch wants a clear line between approved tools, government-configured tools and public ones.
• A departmental AI plan is due. Leadership needs a ranked list of use cases with effort, risk and a first pilot.
• A vendor has proposed an AI platform. The CIO office wants an independent read on data location, model choice and exit terms.
• An existing system predates the June 2025 update. Automated decision systems built before June 24, 2025 had until June 24, 2026 to meet the updated requirements.

What we automate for federal departments and agencies

Correspondence and ministerial briefing drafts. AI sorts incoming correspondence by topic and drafts a first reply from approved lines. An officer edits, and the approval chain stays as it is today.

Application completeness checks. AI reads incoming files, checks them against the checklist and flags gaps before an officer opens the file. The officer makes every decision. This is the pattern the TBS agentic AI guide describes: verifying completeness, retrieving files and escalating borderline cases for full human review. See document processing.

Policy research and consultation summaries. AI groups thousands of consultation comments by theme and links each theme to the source comments. Analysts check the grouping and write the findings.

Bilingual service content. AI drafts French and English versions of web content, FAQs and form guidance for translators to review. Multilingual assistants answer public questions from published content only.

Internal knowledge search. Staff ask questions of policies, directives and past decisions, with answers citing the source document. Access follows existing permissions.

Federal rules that shape the work

Directive on Automated Decision-Making. It applies to automated decision systems in production that make or support an administrative decision about a client, in institutions subject to the Policy on Service and Digital. Before production, the institution must complete, approve and publish an Algorithmic Impact Assessment on the Open Government Portal. For the project, this means we mark every use case as inside or outside the Directive during the audit.

AIA impact levels I to IV. Level I impacts are "little to no, easily reversible, and brief"; Level IV impacts are "very high, irreversible and perpetual". The level sets the review, notice and human involvement required. A Level III design needs more time in the plan.

Guide on the use of generative AI (FASTER). Tools should be Fair, Accountable, Secure, Transparent, Educated and Relevant. Public servants must not input personal information into publicly available generative AI tools. Protected information needs the chief security officer's approval, so tool choice comes before any build.

Guide on the Use of Agentic Artificial Intelligence. TBS recommends tightly scoped internal workflows with limited permissions and accountable public servants making final decisions. Our pilot designs follow that shape.

Official Languages Act. Section 22 requires federal institutions to serve the public in either official language. Public-facing AI output is planned in both languages from day one.

Data location. Azure OpenAI Standard deployments keep processing in the chosen geography, and in Canada that means Canada East for a short list of models. Vendor selection compares these options against your security categorization.

Which package fits

Most federal teams start with the AI Opportunity Audit: it gives a ranked, rule-checked list and a 12-month roadmap. A branch that already has a scoped use case and a cleared tool can go straight to a pilot.

Full scope of each package is on the pricing page.

How the work runs

The engagement follows the Remolda Cycle: Audit → Strategy → Implement → Empower → Evolve.

1. Audit. Interviews with 6–10 people: program leads, front-line officers, the IM/IT team, the privacy office (ATIP) and the security office. We look at process maps, volumes and the tools already approved.
2. Strategy. Use cases ranked by value, effort and Directive exposure. Each one notes the likely AIA questions and the privacy review it would trigger.
3. Implement. A six-week pilot on one internal workflow, run in a government-controlled environment with test or cleared data.
4. Empower. Short training for the officers who use the tool, plus a written AI use policy aligned with the TBS guide.
5. Evolve. A review of results and the next use case.

In our experience the audit is two weeks of work, and calendar time in a department is usually longer. It depends on interview scheduling, security briefings and document clearance. For the compliance side in more depth, see AI compliance in Canada.