The AI Skills Framework for UK Employers: What Every Role Needs and How to Get There

Last updated: 31 March 2026

Why Most AI Training Programmes Are Flying Blind

Ask most HR directors how their organisation is approaching AI skills development and you’ll get one of two answers. Either they’re running an all-staff AI awareness course — usually a generic e-learning module that covers what large language models are and why hallucinations happen — or they’re leaving it entirely to departments to figure out on their own. Neither approach holds up.

The all-staff e-learning approach treats a software analyst, a clinical nurse, a customer service agent, and a finance director as if they have the same AI skills needs. They don’t. The requirements for a frontline admin worker who uses an AI drafting assistant to handle routine correspondence are fundamentally different from those of a procurement manager whose team is evaluating AI contract analysis tools, or a chief technology officer who is accountable for the organisation’s AI governance policy. Training them on the same content at the same depth is wasteful at best and produces a false sense of compliance at worst.

The departmental self-determination approach produces the opposite problem: inconsistency. Some teams develop genuine AI capability while others remain effectively AI-illiterate. Regulatory risk becomes unmanageable because there is no central view of what AI tools are in use, who has been trained on them, or whether that training was adequate.

What both approaches lack is a framework: a structured, role-calibrated view of what AI competency actually means at each level of the organisation, what training achieves each level of competency, and how to assess where each employee currently sits against it. This article provides that framework. It is designed to be practical — something an HR or L&D team can use immediately to audit their current position, prioritise their gaps, and build a coherent training programme — not a theoretical model requiring months of internal validation before anything gets done.

The Five Dimensions of AI Competency

Before mapping skills to roles, it is necessary to define what “AI competency” actually means. There are five distinct dimensions, each of which manifests differently depending on role level. Understanding these dimensions — and recognising that they are not interchangeable — is the conceptual foundation of the framework.

These five dimensions interact but are not substitutable. An employee can have strong AI tool use skills and poor AI governance awareness — a common and dangerous combination. A senior leader can have sophisticated AI governance understanding but almost no practical AI tool use experience, leaving them unable to evaluate what their teams are actually doing. The framework addresses each dimension at each role level, rather than treating “AI competency” as a single spectrum from novice to expert.

Role-Level Competency Matrix

The following matrix maps all five competency dimensions across four role levels: frontline and operational staff, functional specialists, managers and team leads, and senior leaders and executives. For each cell, the description reflects what the competency looks like in practice at that level — not an abstract definition, but observable workplace behaviours and training outcomes.

Table 1: AI Skills Competency Matrix — five dimensions across four role levels. Training route suggestions are indicative; actual requirements depend on role, sector, and AI systems in use. © Training Intelligence (TIQ) Ltd 2026.

The UK Regulatory Overlay

The competency framework above is a practical governance aid, not a statement that every employer has the same legal training duty. The rules that apply depend on matters such as where the organisation and affected people are located, whether personal data is processed, the purpose and risk of the AI system, and the sector in which it is used. Map those facts first and take legal or data-protection advice where the answer is material or unclear.

EU AI Act Article 4 — AI Literacy Obligation

Article 4 of the EU AI Act has applied since 2 February 2025. Providers and deployers within the Act’s territorial scope must take measures, to their best extent, to ensure a sufficient level of AI literacy among relevant staff and other people dealing with AI systems on their behalf. A UK organisation is not automatically in scope merely because it uses AI: test the territorial rules in Article 2, including whether systems or outputs are placed or used in the EU. The European Commission says the approach is flexible and proportionate and that Article 4 does not prescribe a course, certificate, governance structure or documentation format. An AI inventory, role-based needs assessment, learning records and review cycle are useful ways to manage and evidence the work, but they are practical controls rather than a mandated template. National enforcement of Article 4 begins on 2 August 2026.

UK Data Protection Law — Automated Decision-Making

The Data (Use and Access) Act 2025 changed the UK rules for significant decisions made solely by automated processing, and all of its data-protection provisions are now in force. The rules distinguish a solely automated decision — one with no meaningful human involvement — from genuine decision support. Appropriate safeguards can include telling the person about the decision, enabling representations, human intervention and challenge, while special-category data remains more restricted. Employers using AI in recruitment, performance management or other significant decisions should check the current ICO guidance, identify a lawful basis and complete a DPIA where required. Role-based training can help people provide meaningful oversight; it does not by itself make the processing lawful.

Equality Act 2010 — Algorithmic Bias

The Equality Act 2010 applies when an employer or service provider uses an AI system; technology does not remove duties concerning direct and indirect discrimination. Historical or unrepresentative data can create discriminatory outcomes in recruitment, pay, promotion and access to services even without deliberate intent. Training relevant specialists and managers to recognise, test and escalate bias is a sensible risk control. The Act does not, however, create a universal prescribed AI-training course or duration, so combine learning with appropriate testing, human oversight, impact assessment and governance.

Skills England — Digital and AI Skills Priorities

England has several distinct skills programmes, but they should not be blended into a generic promise of funded AI training. The approved Growth and Skills Levy products are apprenticeships, foundation apprenticeships and the specific apprenticeship units listed by Skills England. Skills Bootcamps and the adult digital entitlement have separate eligibility, commissioning and availability rules. AI Skills Boost also points adults towards free partner learning. Check the exact learner, employer, product and start-date rules before treating any route as funded.

How to Assess Your Workforce Against the Framework

A framework is only useful if it connects to a practical assessment process. There are four approaches to assessing your workforce against the AI skills framework described above, each with different coverage, depth, and resource requirements. Most organisations will want to combine at least two.

Self-Assessment Survey

A structured self-assessment survey is the fastest way to get broad baseline coverage across a large workforce. The survey should be organised around the five competency dimensions, with four to six questions per dimension calibrated to the role level of the respondent. For each dimension, questions should ask employees to rate their current confidence on a simple scale (typically 1–5), describe their current actual use of AI tools, and identify specific areas where they feel under-equipped. Confidence ratings are imperfect — the Dunning–Kruger effect means that the most overconfident respondents are often those with the largest actual gaps — so the survey should always include behavioural anchors (“I regularly check AI-generated outputs for factual errors before sharing them” is more useful than “I am confident in my data literacy”). Run the survey with role-level segmentation built in so that you can immediately see where the gaps concentrate.

Manager Observation Indicators

Manager observation provides the behavioural evidence that self-assessment cannot. A simple observation checklist — built around the role-level competency descriptors in the matrix above — gives line managers a structured language for what good AI-competent practice looks like in their team. Indicators might include: does the employee check AI outputs before using them? Do they ask appropriate questions when AI recommendations seem unexpected? Do they disclose AI assistance in line with team policy? Do they use AI tools for appropriate tasks without being prompted? A light-touch version of this checklist can be integrated into the existing supervision or performance review cycle without requiring a separate AI assessment process.

Skills Audit in the Performance Review Cycle

A formal AI skills audit can create a useful documented evidence base. Choose the cadence according to risk and change: an annual review may suit stable use cases, while a new system, role or material incident should trigger an earlier review. Combine self-assessment data, manager observation and a structured development discussion. The output can be a simple profile showing the current level on each dimension, an agreed target and the learning pathway. Where Article 4 applies, these records may help demonstrate the measures taken, although the legislation does not prescribe an annual audit or a particular evidence format.

Using AI Tools to Benchmark — Carefully

There is an irony in using AI to assess AI skills, and it is worth being explicit about the limitations. AI-powered assessment platforms can process self-assessment and behavioural data quickly, identify patterns and generate individual or team-level gap analyses. They also inherit the quality limits of their inputs and may introduce privacy, monitoring, fairness and transparency risks. Use them to generate hypotheses about gaps, then use proportionate human review to validate, document and act on the findings. Complete the appropriate data-protection and employment-law checks before analysing worker data.

Mapping the Framework to UK-Funded Training Routes

One of the practical advantages of a structured AI skills framework is that it makes it straightforward to identify which gaps can be closed through UK government-funded training routes — significantly reducing the net cost of closing them. The following mapping covers the four main levels of the framework.

Foundation: AI Literacy for All Staff

Frontline AI literacy can be delivered internally or through external provision. Free courses are available through AI Skills Boost partners, but employers still need to add their own approved systems, data boundaries, human-review rules and escalation routes. The adult digital entitlement and Skills Bootcamps are separate programmes with their own eligibility and availability; neither should be described as an automatic AI literacy route. Article 4 does not require an accredited course or certificate.

Practitioner Level: Functional Specialists

The functional specialist level may be served by a locally commissioned Skills Bootcamp where the curriculum, learner and employer rules fit. Skills Bootcamps last up to 16 weeks. For an existing employee, the published employer contribution is normally 10% for an organisation with 1–249 employees and 30% for one with 250 or more. Availability and content vary by area and contract.

Do not treat “Growth and Skills Levy short course” as a generic funding category. The live apprenticeship-unit catalogue contains exact approved products. The current AI units are specifically for qualifying employed leaders, while engineering and construction units have different occupational purposes. Match the person to the product page before discussing funding.

Specialist Level: The Level 4 AI and Automation Practitioner Apprenticeship

For a genuine occupational role focused on improving work through AI and automation, check the Level 4 AI and Automation Practitioner apprenticeship. Skills England lists ST1512 version 2.1 with a typical duration of 18 months, 420 minimum compliance hours and maximum funding of £18,000. Version 2.1 applies to starts from 22 May 2026 and uses a revised assessment plan: a project is mandatory and the assessment organisation selects any additional method needed from the published list. Apprentices may be assessed at an appropriate point during the programme. Confirm the assessment specification and funding route before enrolment.

Leadership Level: April 2026 AI Apprenticeship Units

The approved AI leadership products are three standalone Level 5 apprenticeship units: AU0009 for strategy and opportunity, AU0010 for adoption, procurement and governance, and AU0011 for delivery and organisational transformation. Each requires at least 30 eligible live delivery hours and has maximum funding of £750. They are for employed learners aged 19 or over in qualifying leadership roles; they are not automatically added to an Operations Manager or Senior Leader apprenticeship and are not a zero-cost entitlement.

A Practical 30-Day Quick-Start for HR and L&D Teams

The full AI skills framework described in this article is substantial. For HR and L&D teams who need to make immediate progress rather than spending three months on programme design, the following 30-day quick-start provides the minimum viable version of the framework — enough to establish a baseline, identify the highest-priority gaps, and begin closing them through funded training routes.

The 30-day quick-start is not a substitute for the full framework — it is the first sprint in a rolling programme. The key discipline is the 30-day rhythm: each cycle should produce documented evidence of the needs assessment, the training delivered, and the initial outcomes. After three cycles, you will have coverage of your highest-risk role groups and a documented evidence base that stands up to regulatory scrutiny. After six months, you will have the data to build a credible multi-year AI skills roadmap aligned to the organisation’s AI strategy.