ShapedSI Journal · Applied AI
Choosing AI consulting in Canada
An AI consulting engagement should leave you with a clearer decision and a usable next step. For a business in Canada, that starts with the work already happening in the office, shop or field. Bring one real process to the conversation and ask how the proposed change will be reviewed, operated and maintained.
In this guide
01
Bring the process, including the awkward parts
Choose a task people can describe from beginning to end. An enquiry arrives, somebody reads it, information is checked, a response is drafted and the next person takes over. Record where staff wait, copy information or correct the same mistake. Include the exception that happens on a busy afternoon, rather than presenting only the clean version of the process.
Bring a few examples with personal or confidential details removed. Show the forms, spreadsheets and systems involved, who has access and who is responsible for the final decision. That gives a consultant enough context to discuss an approach without needing unrestricted access to your business. A clear process map is useful even if the eventual recommendation is a simpler form or a better handoff.
02
Compare proposals on the same basis
Ask each provider to separate discovery, implementation, licences, integrations, training and ongoing support. A single headline price can hide very different amounts of work. The proposal should name what is included, what depends on access to another system and which changes would require a new scope.
Ask for an example of the acceptance criteria. “An AI assistant for the team” is too open to judge. A more useful target describes the input, the expected draft, the approved source information and the review step. Ask what happens when the system cannot answer, when a record is incomplete or when a supplier changes its interface.
03
Agree on ownership before building
Identify the business owner for the process and the person who will approve the pilot. Confirm who controls the domain, accounts, code, documentation and data exports. Staff should understand which parts they can edit themselves and which require technical support. If you change providers later, the handover should be an understood part of the engagement.
Local availability can make it easier to inspect a workflow with the people who do it. Discuss whether meetings will be in person, by phone or online, and how urgent issues are handled. Do not assume a local address means unlimited support or that a remote provider cannot understand your work. Ask for the working arrangements in writing.
04
Start with a decision you can review
Keep the first pilot narrow enough to assess. Test representative examples, record correction time and ask the staff member reviewing the output whether it is actually easier to use. A convincing demonstration is a starting point; everyday use is the test. Agree beforehand on the result that would justify another stage.
If the pilot is useful, document the operating instructions and the unresolved limits before expanding it. If it is not useful, keep the process map and the findings. Those are still valuable outputs from consulting. The next decision should follow the evidence you collected, rather than the size of the original ambition.
Bring this to the conversation
A working brief.
- One workflow with a named owner
- A few anonymised input examples
- Systems and access requirements
- Separate build and ongoing costs
- Written acceptance criteria and an exit plan
A few useful questions.
Do I need to choose a model before the meeting?
No. Start with the task, data and review requirements. A model choice becomes useful after those are clear.
What should the first engagement deliver?
A defined problem, a proposed approach, the limits of the pilot and a decision about whether the next stage is worth doing.
Original sources
Sources and further reading.
OpenAI’s agent safety documentation describes tool approvals and human review as controls when an agent can use connected tools. Ask a provider how those controls apply to your proposed workflow.
Shape the next step