Do you need a consultant to find out which AI skills your firm lacks? No. The frameworks are published, the funding rules give the skills scan, and the evidence is the work your people do.
How to Run an AI Skills Gap Analysis Before You Buy Training (No Consultant Needed)
By James Cotton · Last updated · Sources checked · 12 min read
Part of our topic guides on AI Skills for Business and Data Literacy.
By James Cotton, Founder, iO-Sphere
What a needs analysis looks for
The CIPD's factsheet on learning needs analysis, dated 20 August 2025, separates "a current or future health check on the skills, talent and capabilities of the organisation" from a training needs analysis, "a one-off isolated event looking at the needs for a specific training activity."
An AI skills gap analysis run before a purchase takes the breadth of the first and points it at one decision. Its question, in the factsheet's words, is "What are the gaps between existing capabilities and new/future requirements". The factsheet lists "Information and analysis from competency frameworks" among the sources, and says the method "applies to smaller organisations too."
The UK's skills gaps are counted nationally by asking employers. In the Employer Skills Survey 2024, 22,712 telephone interviews for the Department for Education and Skills England between June 2024 and January 2025, 88% of employers considered all their employees fully proficient and 12% reported at least one who was not.
The report warns that the survey "can only capture the skills gaps that employers are aware of", and that employers who pay little attention to their staff's skill needs "may be less likely to report skills gaps". Among employers that had not trained in the previous 12 months, the most common reason, given by 69%, was that all their employees were proficient or that there was no reason to train.
The trouble with asking people how confident they are
In 2024, an employer survey run by Ipsos for the Department for Science, Innovation and Technology (DSIT) found 61% of employers with AI-using staff rated them "good" at keeping information safe and private. The companion survey of the public found 19% of full-time workers who had heard of AI (417 people) felt confident in that skill.
The report warns that the two were different studies and are not directly comparable. Both numbers are opinions and neither is a test. Two judgements of one skill sit far apart, and a gap analysis built on either has no way to tell which is right.
In 2024 the same employers rated their AI-using staff good at judging the accuracy and reliability of what AI systems tell them (57%) and at writing clear prompts for generative AI (45%).
Where self-assessment has been tested against performance, the two have come apart. In a study led by Daniela Fernandes at Aalto University, published in Computers in Human Behavior in February 2026, 246 people used AI on 20 reasoning problems from the Law School Admission Test. Their performance rose by three points against a norm population; they overestimated it by four.
Higher AI literacy went with less accurate self-assessment, and a second study of 452 people replicated the findings. A paper for the LAK2026 conference by Zhang and six co-authors found "a low correlation" between school teachers' self-reported and tested AI literacy, with some overestimating and others underestimating. Neither study is of UK workplaces.
The pass mark matters as much as who answers. Lloyds Bank's Consumer Digital Index asked UK adults by telephone in 2024 whether they could do named tasks without help. Read broadly, 82% of the labour force had the essential digital skills for work; read strictly, 48% could do all 20 work tasks.
The index draws on the Department for Education's Essential Digital Skills Framework, last updated in 2019, which set skills out as "I can" statements. When the department measured adults in England for the Survey of Adult Skills 2023 in 2022 and 2023, it tested literacy, numeracy and adaptive problem solving directly on a tablet, and kept self-assessment for social and emotional skills.
A yardstick: eight areas, six levels
The yardstick we use is our AI Capability Framework, agreed in August 2026 and built for the sequence we sell: assess, recommend, teach, measure again. It sits beside the published frameworks and standards in this section, and its area and level names are printed here for you to use without us.
The framework names eight areas in which AI capability shows up at work: tools and context; judgement and evidence; governance and security; ethics and consequence; redesigning work; building and operating; adoption and impact; and adapting.
Every area is scored on the same six levels. Level 0 is not started. Level 1, prompted: does it when shown or asked. Level 2, independent: does it reliably on their own work. Level 3, accountable: does it on work other people depend on and is answerable for the result. Level 4, multiplying: other people's work is better because of them. Level 5, shaping: sets how the organisation does this.
Read up the scale and the axis is who depends on the work. The step from level 2 to level 3 is where someone becomes reliable to an employer, and a confidence question cannot see it, because it turns on what other people do with the result.
The areas map to published references. The AI Skills Framework, in Dr Nisreen Ameen's report AI skills for the UK workforce, co-developed with Skills England and published in October 2025, crosses technical, non-technical, and responsible and ethical skills with entry, mid and managerial job levels. By our count it holds 42 descriptors, and the report presented it as a starting point to be adapted and revisited regularly.
The AI Skills for Business Competency Framework, version 3.0.0, released on 9 March 2026, is led by Dr Matthew Forshaw of The Alan Turing Institute and sponsored by DSIT. Counted the same way, it has 41 duties and 92 competencies, built from knowledge, skills and behaviour (KSB) statements, on six proficiency levels running from Remember to Create, with Evaluate, which the table below uses, at level 5.
The Turing framework marks which KSBs can come from training or self-study and which, "for example, those involving leadership, accountability or stakeholder engagement", "must be acquired through practice in real situations". It does not prescribe how to assess a person; that method comes from the apprenticeship system.
Each apprenticeship standard is a national list of KSBs, and Skills England lists assessing "the prior learning of apprentices at the start of their apprenticeship" among the uses of an occupational standard. Gap analysis is itself a KSB: the Level 4 Business analyst standard includes K19, "The purpose and activities of the gap analysis process", among its 28 knowledge statements.
Its assessment plan gives the employer the job to "decide when the apprentice is working at or above the occupational standard". Ameen's report illustrated its framework with a regional logistics company that used it "to audit existing workforce capabilities, revealing technical and ethical gaps", and a clinic owner mapping a team's roles to entry and mid-level skills; both are illustrations written for the report.
The mapping below is ours: for each of our eight areas, the published reference it lines up with and one statement from that reference.
| Our area, and what it covers | Published reference | One statement from it |
|---|---|---|
| Tools and context: getting a system to do what you want | AI Skills Framework, entry level, technical | "operate embedded AI features (for example, autocomplete, transcription tools)" |
| Judgement and evidence: whether what goes in and what comes out can be trusted, whether it is any use, and how you would know | Turing framework, judging AI output, at its Evaluate level | "Can evaluate AI outputs for accuracy, relevance, fairness, and appropriateness." |
| Governance and security: what the rules require of the work somebody actually does, what an agent may be given access to, and how these systems can be made to leak or be manipulated | AI Skills Framework, entry level, responsible and ethical | "apply data protection guidance" |
| Ethics and consequence: who is affected, and what is right where the rules are silent | Data technician apprenticeship standard, Level 3 | "The ethical use of data, including in relation to its use with Artificial Intelligence and other automated systems" |
| Redesigning work: seeing what should change about how work flows, specifying it, sequencing it, knowing what each change depends on | Turing framework, AI in workflows, at its Create level | "Can combine AI outputs with human expertise to improve efficiency and quality of work." |
| Building and operating: making the thing and keeping it running, up to agents that act on their own | AI Skills Framework, mid level, technical | "use low-code AI platforms for automation" |
| Adoption and impact: getting other people to work differently, and knowing whether it worked | AI Skills Framework, managerial level, non-technical | "evaluate scalability and long-term business value of AI solutions" |
| Adapting: keeping method current; deciding what to take up and retire | AI Skills Framework, mid level, non-technical | "apply new AI tools and support peer learning" |
The method the funding rules already prescribe
Every apprenticeship provider already has to measure anyone aged 19 or over against a list of KSBs. The apprenticeship funding rules for August 2026 to July 2027, from the Department for Work and Pensions, require an initial assessment covering eligibility, prior learning and experience, learning support, and English and maths. Paragraph 27 leaves the tools to the provider.
For apprentices of that age, paragraph 38 has the provider check the person's learning record, discuss the likelihood of relevant prior learning, then conduct "a skills scan against the knowledge, skills and behaviours requirements" of the standard, or of a training plan mapped to it. Diagnostic testing can supplement the scan.
The glossary says the scan must "evaluate and document the extent to which the individual meets each KSB" and "should be validated as accurate by the individual's line manager / employer". The provider holds it as evidence, and records with the employer's agreement where no relevant prior learning turns up.
DWP's guidance on recognising prior learning, last updated in November 2023, says the scan "may be a self-assessment, with a follow-up discussion", in which the provider should "help the apprentice to give an accurate reflection of their skills", talk through "their duties and their experiences", and ask the employer's opinion of the level at which the apprentice is working.
The result carries a price. Paragraph 35 says funds "must not be used to pay for training which will deliver knowledge, skills and behaviours the apprentice already has", so under paragraph 39 what the scan finds comes off the planned hours of off-the-job training (the learning done in paid hours, apart from the person's normal work), and off the price with them.
An apprentice with relevant experience "could spend less time training or qualify earlier", GOV.UK tells employers, and when that apprentice already works for you, the funding split for existing employees applies to the price that remains. On an apprenticeship unit, the shorter funded provision, the unit funding rules make a learner whose existing knowledge and skills overlap 50% or more of the planned hours ineligible for funding.
Paragraph 63 requires the outcome to be shared with the individual and the employer, with "evidence that the employer agrees", and the provider must show the person needs "significant new knowledge, skills and behaviours". The employer's agreement with the provider then sets down that all relevant prior learning and experience has been identified and accounted for in the training plan.
The assessment itself is not off-the-job training (paragraph 84.1), the published minimum hours assume no relevant prior learning (paragraph 85.1), and the rules list initial assessment as an eligible cost (paragraph 106.1).
What employers told the researchers
The workshops behind Ameen's report, held between March and May 2025, found the problem this page answers. Experts "consistently described a lack of clarity from businesses regarding what AI skills they needed", and in larger organisations AI experimentation was sometimes led by innovation teams or department heads "with limited communication to HR or learning and development staff".
The report quotes a workshop participant who works with SMEs on AI planning, in words it says may be paraphrased or composite:
"Employers don't know what AI skills are needed for their business, so the training providers act as consultants, tell me your business plan, your people, and we'll design an AI plan based on that. But many don't have one."
The same chapter found "AI skills" meant fluency with tools such as ChatGPT or Copilot in some workplaces and awareness of bias, data protection and ethics in others, and that some employers hesitated to commission training because of the confusion.
DSIT's 2024 employer survey put numbers on it: 56% of employers using or planning to use AI rated their business's AI knowledge as beginner or novice, and being unsure what training was relevant (50%) was the top barrier to AI training. In spring 2025, DSIT and DBT's adoption research found not having identified a use, and limited AI skills, the two top barriers to adoption.
Four steps to run it yourself
Each step starts from the work, because in our view a request for AI skills is usually a request to get more of a person's own work done well. Our own initial assessment on an apprenticeship follows the same pattern: a skills radar mapped to the standard's knowledge, skills and behaviours, then a professional discussion, for each learner at the start.
- Place each role. Mark the areas that matter to the role and the level its work needs. The AI Skills Framework's prompts help: whether someone uses AI "for your own tasks", helps "embed AI tools within team or departmental workflows", or makes "decisions that shape how AI is adopted across the organisation". That is our axis as a question: who depends on the work?
- Ask for one real task. Have each person show each area on one task from their own week: a draft from Copilot or another tool that they fix or cut, and a source they check. Start from their own account of real tasks, including AI use they taught themselves, then correct it in a discussion about the work; their own account starts the discussion, but the level comes from the work, not from what they say. Leave the questionnaire out.
- Have the line manager validate. The manager confirms each level against what the work needs, as the funding rules ask of a skills scan; DWP's guidance says the employer "needs to support the apprentice to make sure the assessment of prior learning and experience is right". Where the manager and the person disagree, the task decides.
- Tally the gaps, then buy against them. Count gaps by area and level; the employer guide in DWP and Skills England's June 2026 collection put skills needs before training. Whether the route is a custom programme for a team-wide gap or Data & AI Essentials (Level 3), Advanced Data & AI (Level 4) or AI Transformation (Level 4) for one person's duties, buy only against a gap the tally shows.
Frequently asked questions
Is a skills scan the same as a gap analysis?
A skills scan is the narrower of the two. The funding rules define it against one apprenticeship standard, to show "which elements of the standard do not need to be re-taught", inside an initial assessment that "determines the learner's starting point". A gap analysis makes the same comparison for a role or a team against the work it needs, before anyone has chosen a programme.
What does an AI Skills Boost badge show?
Skills England's benchmark, published in January 2026, says anyone who "completes an AI Skills Boost course covering all 6 foundation skills will receive a virtual badge to show they have met this benchmark". The six foundation skills form one bar with no levels. DSIT's explainer counted 1,001,147 AI courses completed, from partners' completion data in January 2026.
How often should we re-run it?
Run it before you buy, again once training is under way, and at three and six months after. DWP's guidance tells providers to check the initial assessment again after an apprenticeship starts, and the 2026 employer guide suggested revisiting training after 3 and 6 months. Repeating the same task check is what turns it into a measure of the training.
What if the gap is at manager level?
The top two levels matter: multiplying, where other people's work is better because of the manager, and shaping, where they set how the organisation does this. The AI Skills Framework's managerial level includes "define accountability for AI use within teams or departments" and "plan investment decisions around AI tools and skills". Score a multiplying manager on the team's work, not only their own.
Upskilling a whole team?
Tailored data and AI training for organisations, from data literacy to technical upskilling, built around your team's real work.