Summary

• More than 44% of surveyed organisations said they use AI tools daily, while training often remains informal.
• The findings reinforce a recognised adoption gap rather than establishing it as a new trend.
• Task-based learning, protected time, governance, and role-specific oversight offer more value than generic tool demonstrations.

AI tools are entering English workplaces faster than organisations are building the structured skills, supervision, and governance needed to use them consistently, according to a new package of employer research and guidance.

Skills England and the Department for Work and Pensions found that more than 44% of surveyed organisations use AI tools daily, although many remain at an early stage of adoption. Employees frequently learn through experimentation, colleagues, online videos, and prompts embedded in the software.

The research draws on 23 workshops involving around 150 organisations, ten case studies, and a survey of 536 employers. Formal courses, employer-led programmes, and informal learning were examined across organisations of different sizes and sectors.

An accompanying employer guide recommends training built around real tasks and decisions, combining technical capability with responsible use while helping staff recognise situations in which AI should not be used.

Informal learning spreads uneven practice

Employees rarely wait for a formal curriculum before testing software that can draft an email, summarise a document, produce code, or reorganise information. Such experiments can identify useful applications quickly, particularly when teams understand the problems in their own processes better than a central training function.

Learning through trial and error also distributes good practice unevenly. One employee may verify outputs and avoid confidential information, while another pastes sensitive material into an unapproved service or relies on an answer that sounds credible but is wrong.

The resulting exposure extends beyond technical mistakes because AI can affect professional advice, recruitment, customer communication, intellectual property, and regulated decisions. Problems may remain hidden while outputs are used internally, only becoming visible after material reaches a customer, regulator, or member of the public.

The findings reinforce an established pattern in enterprise AI adoption rather than revealing it for the first time. Access to generative tools has expanded quickly across many organisations, while policies, operating models, and performance measurement have developed more slowly.

Skills England adds detailed English evidence about how that gap appears inside day-to-day work. Its PRIMES framework says training should be practical, reachable, integrated, modular, expandable, and sustainable, placing particular emphasis on learning within the processes employees actually perform.

Roles carry different consequences

A generic introduction to prompting may establish basic literacy, but it cannot address the different risks attached to AI use in finance, care, engineering, law, sales, manufacturing, or public administration.

An employee using AI to produce an internal first draft requires different controls from someone relying on it to recommend credit, screen candidates, interpret clinical information, or operate industrial equipment. Higher-consequence work requires evidence about validation, escalation, record keeping, and human accountability rather than more elaborate prompting alone.

Managers need a separate capability because they decide where AI belongs in a process, how outputs will be checked, whether productivity gains are genuine, and whether time saved in one stage reappears as additional review elsewhere.

Without that management layer, employees can become faster at producing material that an organisation cannot safely use. A quicker first draft has limited value when specialists must spend longer finding unsupported statements, checking confidential data, or rebuilding the work from original sources.

Skills England’s case studies suggest that larger employers including Airbus, KPMG, and Roche are developing structured pathways and governance, while smaller organisations rely more heavily on applied learning and peer support. SMEs may lack dedicated training departments, but they can still define approved tools, build short role-based modules, and reserve time for controlled practice.

Training needs time and measurement

The research identifies workload, cost, staff pressure, unclear provision, and fear of technical failure among the barriers to upskilling. Those are organisational constraints that another optional online course will not resolve.

Protected learning time forms part of implementation because employees expected to acquire new capabilities alongside an unchanged workload are more likely to take shortcuts. Expertise can also become concentrated among people already comfortable experimenting, widening differences between office-based staff with easy tool access and operational workers receiving less investment.

Course completion provides only a weak measure of readiness. Organisations can instead assess whether employees choose suitable tools, recognise prohibited data, verify outputs, escalate uncertainty, and produce work meeting the standards already applied to the role.

Workflow measurement should include review and correction. A system may reduce drafting time while creating additional checking, or move effort from a junior employee to a more expensive specialist whose approval is required before the output can be used.

The evidence has limits because the survey is self-reported, the published guidance applies to England, and daily use does not indicate how intensively a tool is deployed. An organisation where a small number of employees use AI each day differs from one whose core operations depend on it.

Even with those caveats, the research provides a useful distinction between access and capability. Employees are already learning inside workflows; whether those experiments become a controlled source of improvement will depend on task design, management, training time, and the quality of oversight surrounding the tools.