AIforBC practical guide
How to train a team to use AI at work
Teach one shared method, practise on approved real tasks, require source and privacy habits, and measure whether work improves over 30 days.
Short answer
AI training works when people practise on real work inside clear boundaries.
A useful program combines approved tools, permitted-data rules, a shared prompting method, source checking, human-review expectations, role-specific exercises, and a 30-day practice period. Measure changed behaviour and work quality—not attendance alone.
Why a single “AI 101” session usually fades
Employees may enjoy a demonstration and still return to their old process the next day. The gap is not enthusiasm; it is operating design. People need to know which tool to use, what information is allowed, how to frame the task, what a good result looks like, who reviews it, and where the output belongs in the workflow.
Training should therefore produce reusable work instructions and habits, not only awareness.
Before training: establish five boundaries
- Approved tools: name the accounts and configurations employees may use.
- Permitted information: explain what is public, internal, confidential, personal, or prohibited.
- Human responsibility: identify outputs that always need accountable review.
- Escalation: give employees a person or channel for uncertain cases and incidents.
- Recordkeeping: define when sources, prompts, approvals, or versions must be retained.
These boundaries should be short enough to use. A policy nobody can apply during work will not control behaviour.
Teach one repeatable method
Instead of distributing dozens of prompt tricks, use a shared sequence:
Task
State the work to be completed and the decision it supports.
Context
Supply relevant facts, audience, examples, definitions, and source material.
Constraints
Specify permitted sources, format, tone, exclusions, and privacy boundaries.
Quality test
Explain how the output will be checked and what must be escalated.
Improve
Review, correct, compare, and turn the useful approach into a reusable instruction.
A four-week adoption plan
| Week | Focus | Evidence |
|---|---|---|
| 1 | Approved tools, data boundaries, core prompting, and failure modes | Each learner completes one low-risk task and explains how it was checked. |
| 2 | Role-specific documents, research, communication, or analysis | Each learner records a baseline and one supervised workflow attempt. |
| 3 | Repeatability, templates, peer review, and exceptions | The team turns useful attempts into shared instructions and records corrections. |
| 4 | Evaluation and next decision | Managers compare time, quality, adoption, review effort, and incidents with the baseline. |
Choose exercises by role
- Administration: prepare meeting briefs, organize information, and draft routine communication.
- Sales: research accounts, prepare call notes, and improve follow-up while respecting consent and marketing rules.
- Operations: summarize logs, compare procedures, classify requests, and draft work instructions.
- Managers: challenge assumptions, compare options, create decision briefs, and review team use.
- Job seekers and professionals: research roles, improve application material without inventing experience, practise interviews, and build role-specific workflows.
Use de-identified or approved information during training. Higher-consequence exercises should remain simulations until the controls and review process are ready.
Measure capability, adoption, and outcomes separately
Capability asks whether people can perform the method. Adoption asks whether they use it in approved work. Outcome asks whether the workflow became faster, better, safer, or more valuable.
| Layer | Example measure |
|---|---|
| Capability | Can the learner provide context, request sources, identify uncertainty, and review the result? |
| Adoption | How many approved workflows were used at least twice during the practice period? |
| Quality | What proportion of outputs passed human review without material correction? |
| Efficiency | How did completion time change after including review and correction time? |
| Risk | Were prohibited data, unsupported claims, or missed approvals identified? |
The manager’s decision after 30 days
Expand only the workflows that show useful evidence and manageable risk. Revise tasks where review effort erases the benefit. Stop uses that depend on prohibited information, unreliable outputs, unclear accountability, or employee workarounds.
People should leave knowing both how to use AI effectively and when not to use it.
Source trail
Primary sources used in this guide
- Compendium of best practices for human-centered AI in the world of workEmployment and Social Development Canada
G7 best practices on skills development, privacy, fairness, safety, transparency, accountability, and worker involvement.
- Toolkit for SMEs deploying artificial intelligenceInnovation, Science and Economic Development Canada
Practical guidance for secure, responsible, and trustworthy AI adoption by small and medium-sized businesses.
- AI, privacy, and your businessOffice of the Privacy Commissioner of Canada
Privacy principles relevant to staff use of generative AI systems.
- Guide on the use of generative artificial intelligenceGovernment of Canada
Public-sector guidance on cautious experimentation, risk evaluation, and limiting use to cases where risks can be managed.
AIforBC uses official and primary sources where practical. This guide provides general operational education, not legal, privacy, cybersecurity, or financial advice.
Need help choosing?