Funding AI transformation with the apprenticeship levy: a 2026 UK employer's guide

Most "AI training" programmes deliver tool tutorials and call it transformation. They get teams comfortable with a few tools, the budget gets spent, the org chart looks the same six months later. If you're an employer with a levy pot and a real AI transformation question, this guide is how the apprenticeship route actually fits.

What changed in 2026, and why it matters here

The bodies you are told to check have been reorganised. The Institute for Apprenticeships and Technical Education no longer exists; its functions sit with Skills England. The ESFA closed in March 2025 and its functions moved into the Department for Education. If a provider is still citing IfATE approval or ESFA contracts in their sales material, that is a small but real signal about how current their compliance thinking is.

Ofsted changed how it inspects. From November 2025 the renewed framework replaced a single overall grade with separate judgements on achievement, curriculum and teaching, participation and development, inclusion, and leadership and governance, graded Exceptional, Strong standard, Expected standard, Needs attention or Urgent improvement. Ofsted states plainly that there is no read-across to the old Outstanding to Inadequate scale. Two practical consequences: a provider quoting an "Outstanding" from 2019 is quoting something that no longer maps onto how they would be judged today, and grades now differ by provision type, so a provider can be Expected standard for apprenticeships while scoring differently on adult learning. Ask which judgement applies to the apprenticeship provision specifically.

Nobody can show you an achievement rate for this standard yet, and anyone who implies otherwise is selling

This is the single most important thing to understand before you buy. There are currently 121 approved providers delivering the Artificial Intelligence (AI) and Automation Practitioner standard at Level 4, with a funding band of £18,000. The number of them with a published achievement rate for it is zero.

That is not a gap in our data. It is the standard being new enough that no cohort has run far enough through to appear in the Department for Education's qualification achievement rate figures. Achievement rates measure the proportion of starters who complete and pass, so they cannot exist until people have finished.

For most apprenticeships we would tell you to start with the achievement rate, because the spread between providers is wide and it is the best single predictor of whether your apprentice finishes. Here that is not available to anyone, which means every provider is competing on claims rather than results. Some will quote their achievement rate on other standards. That is reasonable evidence about the organisation and no evidence at all about this programme.

What to substitute for the missing number

• Their achievement rate across everything else, with cohort sizes. A provider running 80% across two thousand apprentices has shown it can deliver; one at 100% across nine has shown very little. Our tables carry cohort sizes for exactly this reason.
• Their most recent Ofsted judgement on apprenticeship provision, and the date. Under the renewed framework, ask which evaluation areas were graded and how.
• Who actually teaches it. For a standard this new, the difference between providers is usually whether the people delivering it have built anything with these tools or have adapted a digital curriculum they already owned.
• What the end-point assessment looks like, and whether they have taken anyone through it yet on this standard. Many have not.

Why use this site for this decision

An honest declaration, because you should know who is telling you this. I built Find a Training Provider, and I also run businesses that use apprenticeship funding and advise employers on designing programmes around it. That is a commercial interest, so here is how the site is built to stop it distorting what you see.

The league tables come from the DfE's published achievement data and nothing else. No provider can pay to appear in them or to move up them, and none ever will. Where a provider has paid for a featured placement in search results, it says so on the row. Where we hold no data, we say we hold no data rather than filling the gap.

What we add beyond the numbers is the part this page is doing now: telling you when the number you would normally rely on does not exist, and what to ask instead. If you want a second pair of eyes on a shortlist or on how a cohort is being designed, that is what the enquiry form is for, and it costs nothing.

The reframe: transformation is the goal, training is the funding mechanism

The honest version of the question most employers ask is not "how do we train our team in AI." It is "how do we redesign how this business runs, given that AI exists." Those are different problems with different price tags and different success criteria.

Adoption training answers the first question: people get certified in a tool, finish a module, take a screenshot for LinkedIn. Real transformation answers the second: workflows change, headcount-per-output ratios shift, the operating model adjusts.

If you're spending levy money on the first when you needed the second, you have not failed at AI. You have answered the wrong question. The apprenticeship route can fund either, but you need to know which one you're buying.

What the apprenticeship levy can fund

The UK apprenticeship levy is a payroll tax of 0.5% on UK employers with an annual pay bill above £3 million. Levy-paying employers draw down their fund to pay for apprenticeship training delivered by approved providers. Non-levy employers (the majority of UK businesses) can still access apprenticeships via 5% co-investment, with the government covering the remaining 95%, and very small employers taking on 16-18-year-old apprentices pay nothing at all.

What the levy buys you that off-the-shelf training cannot:

• Funded full-time learning while employed. Apprentices stay in role, do real work, learn against a structured standard, and you do not pay their training costs out of operating budget.
• A multi-month learning arc, not a weekend bootcamp. Standards run 12-24 months typically. Genuine skill compounding happens, not just course completion.
• Assessment against an industry-recognised standard. External validation of competence, not just a vendor certificate.
• Optionality on cohort design. You can put any practical number of people on the same standard at the same time. Five to fifteen is a common starting size, but cohorts can be much larger when the demand is there. Build a programme around your business and use the cohort as a transformation vehicle rather than a training delivery.

That last one is where the real leverage sits, and where most employers leave value on the table.

The AI standard that matters: AI Level 4

The most directly relevant standard for AI work is the Level 4 Artificial intelligence (AI) and automation practitioner, known across the sector as AI L4 (the formal Skills England reference is ST1512). It carries a maximum funding band of £18,000 per learner and is delivered by 58 registered providers currently listed on FATP across the UK. It covers practical AI capability: prompt design, automation workflow build, output evaluation, ethics and governance, integration with business systems.

What AI L4 is good for: building a meaningful base of AI-capable people inside your business who can identify, design, and build workflow improvements with current AI tools.

What AI L4 is not: a transformation strategy. It is a vehicle. What you choose to do with the cohort, what real workflow problems they tackle, who their internal sponsor is, what the cohort produces, that is the strategy, and the standard does not write it for you.

And that's the honest read on most providers in this market: they are training providers, not transformation partners. They will deliver the standard competently and stop at the standard's edge. If you need the strategy and transformation layer wrapped around the cohort, the people designing it should sit outside the delivery contract. Alex Lockey works with employers in exactly that gap, architecting the programme around real workflow problems and orchestrating the apprenticeship provider to deliver against it.

You can browse the standard's full provider list on FATP here, including achievement rates, employer satisfaction scores, and regional coverage.

Why a cohort beats one-off learners

Most employers default to enrolling individual apprentices into whichever standard fits their role. That works at small scale. It does not transform anything.

A cohort treats AI uptake as a coordinated programme rather than a scattered set of personal-development plans. A meaningful group on the same standard, starting at the same time, working on the same defined business problems, with a single delivery partner and a single internal sponsor. Five to fifteen learners is a comfortable starting size; much larger cohorts are absolutely viable where the demand and the internal scaffolding support it.

What that buys you:

• Peer learning compounds. Cohort members debug each other's prompts, share workflows, and build internal pattern libraries that one-off learners never produce.
• Programme design can be tailored to your business. With one learner, the provider runs their standard curriculum. With a cohort, the curriculum can be sequenced around your actual operating problems.
• Internal sponsorship survives. Cohorts produce visible outputs that justify the next investment. Individual apprentices often graduate into a business that has forgotten why they were enrolled.
• Tender leverage. A 10-learner cohort is a real piece of business to a training provider. You can run a structured tender, score providers against criteria you actually care about, and pick on quality rather than on whoever replied first.

This last point is where things have shifted recently. The UK apprenticeship sector is in flux: Skills England replaced IfATE during 2024-2025, the August 2026 subcontracting rules are tightening, and several providers we work with through FATP have flagged operational stress. Running a tender for cohort delivery, with explicit scored questions on subcontracting compliance and delivery stability, has become a meaningfully better mechanic than picking a partner cold.

Picking a delivery partner: what actually matters

The 58 AI L4 providers listed on FATP vary wildly. Some are large nationals scaling fast and not always cleanly. Some are specialists doing genuinely thoughtful AI work with small cohorts. Some added AI L4 because their existing digital learners can sit alongside the new standard at low marginal cost, with little real AI delivery experience underneath, or indeed any internal AI expertise at all. Some offers are wafer thin.

What to look for, in order of importance:

1. Actual AI delivery, not "we updated our digital curriculum." Ask to see a project an existing apprentice produced. If they can't show one, they have not delivered a cohort yet, regardless of what their website says.
2. Achievement rate on adjacent standards. AI L4 is new enough that few providers have multi-year completion data on this specific standard. Their achievement rate on related Level 4 digital standards (Data Analyst, Cyber Security Technician, Software Developer) is a fair proxy. Above 65% is solid, above 75% is strong.
3. Coach quality, not platform features. Ask who specifically would be your cohort's lead coach, their CV, and whether you can speak to them. A great provider with a weak coach for your cohort is a weak cohort for you. You're looking for coaches who have actually implemented AI inside a business, not someone who took the same course last month that they're teaching you this month.
4. Off-the-job time model that fits your operating reality. The 20% off-the-job rule has been revised under Skills England and varies by standard. Cohort delivery typically uses block release, weekly day-off, or distributed self-study. Pick a model that fits how your team works.
5. Willingness to design around your business problems. If their answer to "can we build the cohort around our actual workflow priorities" is "we deliver our standard curriculum, that's what's funded" then you have your answer about whether this cohort will transform anything.

Provider achievement and satisfaction data lives on each provider's FATP profile. The top-rated providers ranking indexes them quarterly on achievement rate, satisfaction, scale, and responsiveness. The digital apprenticeships ranking is the closest sector facet for AI L4 today.

The harder question: who designs your programme?

Most providers will sell you their standard curriculum delivered through their standard delivery model. That is the default purchase. It funds compliantly, lands learners through the gateway, hits achievement rate. It rarely transforms anything.

The alternative is to treat the cohort as a transformation programme that happens to be delivered under the apprenticeship framework. That means someone needs to design the programme: which business problems the cohort works on, in what sequence, with what success criteria, mapped to which parts of the standard's knowledge-skills-behaviours so the funded delivery still flies. Most providers cannot do this, or will not for the price they're being paid.

The practical options for employers:

• Find a provider that does this natively. A small number do. They tend to be specialist, smaller, and book up early.
• Design the programme yourself. Realistic if you have a senior internal sponsor with operating experience, AI fluency, and the time. Most L&D functions do not.
• Bring in a programme architect alongside the provider. Someone who designs the programme, scopes the workflow problems, sets the success criteria, and works with the provider on delivery. The architect is funded outside the levy; the delivery is funded by the levy.

The third option is what we are increasingly seeing employers adopt, particularly where the cohort is meant to drive a meaningful operating-model change rather than just upskill a few individuals. For employers who want this layer built in alongside the cohort delivery, we partner with Alex Lockey, an AI strategy and transformation specialist who designs and architects the programme, scopes the workflow problems, and works with the chosen provider on delivery.

Where to start

The decisions, in order:

1. Is this a training problem or a transformation problem? If training: enrol individuals or small clusters into the standard with a competent provider, you're done. If transformation: keep reading.
2. What is the named business problem you want the cohort to work on? Not "improve AI capability." Something like "reduce time-to-close on regulatory filings by 50%" or "build a cohort of internal AI champions across our four functions who each launch one production workflow inside 9 months." If you can't name it, the cohort cannot be designed around it.
3. Cohort sizing. Size it to the problem and the appetite. Five to fifteen learners is a manageable starting cohort with strong peer dynamics. Much larger cohorts work too, particularly where you're rolling out across multiple functions or sites and want consistent uplift at scale. Below five is usually hard to justify economically for a serious provider.
4. Run a tender. Issue a structured invitation to 4-6 providers, score against criteria you care about including August-2026 subcontracting compliance, pick on weighted score rather than vibes.
5. Decide on programme architecture. Either the provider handles it natively, or you bring in an architect to design and oversee.

Register interest, or get programme design help

If you're at step 1 or 2 above and thinking about an AI L4 cohort starting later in 2026, we're tracking interest from UK employers to coordinate cohorts and connect you to suitable delivery partners.

Register your interest in an upcoming AI L4 cohort and we'll send you the cohort opportunities as they form. No obligation, no sales pitch.

If you're at step 4 or 5 and need help designing the programme itself, that's a different conversation. The author of this guide, Alex Lockey, designs and architects AI transformation programmes for UK employers and orchestrates the delivery through apprenticeship providers. You can read more about how he works at alexlockey.com.

Further reading on FATP

• The AI L4 standard page — full provider list, achievement rates, regional coverage
• Top digital apprenticeship providers — quarterly ranking, the closest sector facet for AI L4
• The best apprenticeship training providers in the UK, 2026 — broader guide on how we rank providers
• Editorial methodology — how FATP sources, verifies, and updates its data