AI training for employees lags far behind AI use

Most of your staff already use AI, and most learned it without help. In Microsoft and LinkedIn's 2024 survey of 31,000 knowledge workers, only 39% of AI users had received AI training from their company¹. Good training teaches people to check output and to notice when AI is out of its depth, because controlled studies show that's where experienced professionals go wrong. Small firms train least².

Shadow AI at work is already common

Gallup finds 52% of U.S. employees use AI at work at least a few times a year, and 15% use it daily³. In Canada, 51% of employees told KPMG they use generative AI at work, up from 46% in 2024⁴.

Bring-your-own use is higher at small and medium-sized companies, where 80% of AI users brought their own tools¹. BCG found 54% of workers would use AI tools even if their employer hadn't authorized them⁶. Only 29% of Canadian employees told KPMG their employer has a full policy on acceptable uses, so most staff decide for themselves what goes into a prompt⁴.

Few staff feel well trained, and Canada trails

BCG's 2026 survey found 36% of respondents feel properly trained, the same share as in 2025, though it covers only large companies outside Canada⁷. Among Canadian employees who use generative AI at work, 48% told KPMG their employer offers enough training to use it well⁴. In KPMG and the University of Melbourne's 2025 study across 47 countries, 24% of Canadians said they'd had AI training against 39% globally, which ranked Canada 44th⁸.

The AI skills gap is widest in small firms

In spring 2026, 19.2% of Canadian businesses had used AI to produce goods or deliver services in the previous year, triple the 2024 share⁹. When Statistics Canada asked those businesses what they'd changed because of AI, training existing staff came first at 32.0%, and the share climbs with company size².

The smallest firms were also the most likely to change nothing. Among AI users with 1 to 4 employees, 62.4% made none of the four staffing changes Statistics Canada asked about, against 28.3% at 100 or more².

Employer support goes with more AI use

Economists Anders Humlum and Emilie Vestergaard surveyed about 25,000 Danish workers in 11 AI-exposed occupations¹⁰. Where employers took no active steps, about 41% had used AI chatbots at work. Where employers combined encouragement with business tools and training, 93% had¹⁰.

Gallup sees the same link in the U.S.: employees who strongly agree their manager backs AI use are 1.7 times as likely to use it a few times a week or more³. BCG found people who got five or more hours of training, especially in person with coaching, were significantly more likely to become regular users, though it didn't publish the size of that effect⁶.

These are associations. Employers that train may also buy better tools and manage change well, so none of these studies shows that training alone caused the higher use.

AI skills pay more in job postings

Lightcast analyzed more than 1.3 billion job postings and found those listing AI skills offered 28% higher salaries, nearly $18,000 a year more¹¹. PwC's 2026 Global AI Jobs Barometer puts the average wage premium for AI skills at 62%¹².

What AI training for employees should cover

Three controlled studies measured what happens to accuracy or learning when people work with AI. Each one points to something a course should teach.

The consulting result is a warning for courses that only teach prompting, since the group given a prompting overview did worst on the task the model couldn't handle. The productivity research covers the rest of that study. In Anthropic's trial, the biggest gap between groups was on debugging questions¹⁵.

Workers admit the checking gap: 58% in KPMG's U.S. study said they rely on AI output without properly evaluating it⁵. AI users in Microsoft's 2026 survey of 20,000 ranked quality control of AI output as the most important human skill, at 50%¹⁶. A course built on these findings asks staff to:

• Check facts, figures, quotes and sources against the original before anything reaches a client.
• Practise on their own tasks where the model gets it wrong, so they know what a confident wrong answer looks like.
• Agree when not to use AI, such as for judgment calls or output nobody can check.
• Ask the tool to explain its reasoning, which matters most for juniors still building skills.
• Keep confidential data out of any tool the firm hasn't approved.

What the evidence doesn't show yet

We couldn't find a randomized trial that isolates the effect of a workplace AI training program on productivity or pay. In the positive results, access and training arrived together. Large employers have grown less sure too: in Wharton's 2025 survey of senior leaders at big U.S. companies, confidence in training as the main path to AI fluency fell 14 points from 2024¹⁷. PwC found AI-exposed entry-level roles are seven times more likely to ask for senior skills such as judgment, which is a reason to teach judgment early¹².

Where to start at a 10 to 500 person firm

1. Find out which AI tools people use, on which accounts and for what. Make it safe to answer, since 52% of AI users in Microsoft and LinkedIn's survey were reluctant to admit using AI for their most important tasks¹.
2. Write one page of rules: approved tools, the data that never goes into a prompt, and who reviews AI output before a customer sees it. How AI knowledge assistants work explains why permissions matter.
3. Move work onto approved business accounts, so people train on the tools they'll use every day.
4. Plan five or more hours of hands-on training per person on real tasks, in person with coaching where you can, the pattern BCG links to regular use⁶.
5. Teach checking before prompting, and ask managers to use AI in front of their teams, since manager support goes with more use in Gallup's data³.
6. Pick one task each team repeats weekly, time it and count its errors before training, then measure both again a month later.

For a baseline before you start, the readiness check scores a business on goals, people, workflows, tools and rules in ten questions.