Put the strategy in front of a real project
The federal plan is strongest when it is read as an operating test, not a list of approved technology. A department still owns the decision, the records and the harm even when its model or vendor changes.
A fictional public-guidance assistant shows how to connect the four strategy priorities to scope, training, evidence and a release decision.
Write a bounded project brief before a public-service AI demonstration expands into client data or administrative decisions.
Keep in mind: The strategy describes a direction for government. It does not prove that a particular department or system follows it.
In this article6 sections
Editorial note: This project test is an editorial interpretation of the federal strategy and related guidance. It does not certify a department, system or procurement and should not replace the rules that apply to a real public service decision.
The strategy does not approve your pilot
Canada’s federal AI strategy for 2025–2027 sets priorities for government adoption. A team still needs to establish whether its particular use is authorized, suitable and ready. Buying a tool named in a presentation or keeping an employee in the process does not settle those questions.
Start with the service problem. If staff spend time locating a public policy paragraph, a search aid may be enough. If a system ranks client files, its effect reaches the administrative decision itself. The same model could sit inside both workflows, but the evidence required to justify each would differ. Our project brief makes those differences visible before a demonstration turns into a commitment.
Four priorities, four things to put in the brief
The strategy identifies central AI capacity; policy, legislation and governance; talent and training; and engagement, transparency and value to Canadians. We translate those priorities into questions a project sponsor can answer. The right-hand column is our proposed evidence, not a new Treasury Board requirement.
Scroll the table sideways to see every column.
| Priority | Project question | Proposed evidence |
|---|---|---|
| Central capacity | Which shared support or infrastructure will this use? | Named service owner and confirmed availability. |
| Governance | Who can approve this use and stop it? | Scope decision, relevant reviews and accountable owner. |
| Talent and training | Can staff detect and handle its likely mistakes? | Role-specific exercise with a documented answer key. |
| Public value | What improves for the person receiving the service? | Baseline, error measure and accessible correction route. |
Three versions of a document assistant
A public-information assistant retrieves passages from published guidance. Its basic test is whether an answer matches the correct, current paragraph and explains when the source does not answer the question. A client-file summarizer adds personal information and the risk of omitting something consequential. A priority-ranking tool adds a further question: does its output change who receives attention first?
Do not let a pilot’s original description survive after its role expands. Moving from public lookup to client summarization is a data change. Moving from summary to ranking is a decision change. Each should trigger a fresh examination of authority, risk and evidence. The federal scope guide says that even a human's final decision can be within the directive when a system assists judgment; a small pilot on real clients may count as production. An experiment using only invented records is a different situation. The strategy alone cannot answer the applicability question.
federal Algorithmic Impact Assessment reading guideCheck the scope question when a document assistant starts influencing an administrative decision.
A filled brief for a deliberately narrow pilot
This fictional pilot helps staff find the relevant paragraph in public service guidance. It cannot access client files or write to a case-management system. Its narrow boundaries are intentional: the team can test whether retrieval helps before taking on decisions or personal data.
Scroll the table sideways to see every column.
| Field | Entry |
|---|---|
| Problem | Staff must locate the current public guidance for a stated question. |
| Permitted output | Document title, section, passage and an explanation of any missing answer. |
| Excluded use | Client assessment, eligibility ranking and changes to records. |
| Baseline | Time and errors when staff use the existing search process. |
| Stop condition | Restricted material appears, or a material instruction is unsupported. |
| Review owner | A named service owner must be assigned before testing with staff. |
| Evidence retained | Source version, test question, selected passage, correction and elapsed review time. |
Train for the mistake, not the interface
A demonstration often shows where to type. Training should also show what to reject. Give staff two conflicting versions of a policy and ask which applies to a dated case. Include a question with no answer in the supplied material. Ask them to explain the escalation path without relying on the assistant to invent one.
This is an instructional design proposal, not evidence that any department has delivered that training. Its benefit is inspectability: a manager can see whether the exercise covers the failure the pilot is likely to encounter. An attendance count cannot establish that staff can identify the wrong version.
Decide what would justify continuing
Compare the pilot with the current process using the same kinds of questions. Count corrections and the time needed to verify an answer. Keep service accessibility and the ability to reach a person in the review; a faster internal summary can still make the public experience worse.
At the decision meeting, record proceed, revise or stop, with the unresolved conditions and their owners. If the sponsor cannot identify who has authority to stop the pilot, the brief is unfinished. A department’s implementation evidence, rather than the national strategy’s ambition, should support the next step.
Continue with the original sources
These claim-relevant primary and first-party references support the reporting above. Open them for technical detail, current requirements and subsequent updates.
- canada.caAI Strategy for the Federal Public Service 2025-2027 ↗The strategy's full text, used for its scope, four priority areas and stated implementation direction; a strategy is not evidence that every action is complete.
- canada.caGovernment of Canada guide on generative AI ↗Operational federal guidance used for the responsibilities and cautions applied to generative-AI use by public servants.
- canada.caTreasury Board: guide on the scope of the automated decision directive ↗Official scope guidance explains partial automation, human involvement and the production boundary. The directive itself still governs applicability.
Finished reading? Save that here without waiting for a timer.
September 23: strengthened the directive scope explanation with the official guide and replaced a source URL inaccessible in this review.
See something we should fix or clarify? Read the corrections policy or tell the newsroom. Material changes are noted here.
