Growth & Skills Levy for AI Training: How to Fund It

By James Cotton, Founder, iO-Sphere

Most employers ask the wrong question about the levy. They ask how to stop the funds expiring. The better question, the one that actually builds an AI-capable workforce, is what capability you'd buy with that budget if it were cash. Because it is.

The Growth & Skills Levy (formerly the Apprenticeship Levy) is a capability budget with a shelf life, and AI training is squarely within what it can fund. Here is our flat position, defended below: the funded route you can use today at scale is Level 4, and the transfer rule is, in our experience, the most underused line in the whole system.

Key figures at a glance

Levy rate: 0.5% of an employer's annual pay bill, for employers with a pay bill over £3 million (HMRC PAYE, unchanged since 2017)
Annual allowance: £15,000, offsetting the 0.5% charge (unchanged since 2017)
Data analyst (L4) funding band: £15,000 max (Skills England, ST0118, 2021)
Levy transfer: Up to 50% of unused funds to another employer (raised from 25% on 22 April 2024, gov.uk)
Co-investment (starts from 1 Aug 2026): £0, free to the employer, for any apprentice aged 16 to 24 at the start, at every employer type including a levy payer whose pot has run dry. For a 25+ apprentice: 5% at a non-levy employer, 25% at a levy payer with insufficient funds (gov.uk apprenticeship funding rules 2026-27, v3, amended 29 July 2026)
Enterprise AI pilots with no P&L return: 95% (MIT NANDA, "The GenAI Divide: State of AI in Business 2025", fieldwork January to June 2025)

What is the Growth & Skills Levy?

The Growth & Skills Levy (formerly the Apprenticeship Levy) is a UK tax of 0.5% of an employer's annual pay bill, paid by employers whose pay bill exceeds £3 million a year. HMRC collects it monthly through PAYE, and a £15,000 annual allowance offsets the charge: so in practice you pay the levy on the portion of your pay bill above £3 million. If you're a levy payer in England, the funds land in your apprenticeship service account, ready to spend on approved training. The underlying mechanic hasn't changed since the levy was introduced in 2017. The rate, the £3m threshold and the £15,000 allowance are all the same.

One point worth clearing up early: the levy is collected UK-wide, but the account you spend from only covers apprentices who live in England. Scotland, Wales and Northern Ireland run their own apprenticeship systems, funded through the block grant. An employer with staff across the UK pays on the whole pay bill but can only spend the English share through the service account.

How the Growth & Skills Levy differs from the old Apprenticeship Levy

The Growth & Skills Levy works the same way the Apprenticeship Levy did, and from April 2026 it also starts funding shorter, more flexible training alongside full apprenticeships. The rename is the visible change; the substance is the added flexibility. The first of that flexible training arrives from April 2026, beginning with short apprenticeship units deliverable from 28 April 2026. This is an ongoing transition, not a single-day switch: the 2026-27 funding rules still describe the transformation as in progress.

Two administrative shifts sit alongside the rename, and it's worth getting the dates precise because they're easy to conflate. Policy responsibility for apprenticeships and adult skills moved to the Department for Work and Pensions (DWP) on 16 September 2025 (confirmed by written ministerial statement HCWS930); higher education and under-19 skills stayed with the Department for Education (DfE).

Ownership of the funding-rules guidance document itself then formally moved to DWP on 1 April 2026: the 2026-27 rules are the first DWP published. Apprenticeship standards, including the data and AI ones, are maintained by Skills England.

If you're briefing a provider or a board, the practical translation is this: nothing about how you access your funds has changed, but you now have a wider set of things you can spend them on, and some funding mechanics tighten from 1 August 2026 (covered under the step-by-step and pitfalls below).

Which AI training currently qualifies for levy funding, and which routes are traps

The clearest, most durable way to fund AI capability through the levy today is a full apprenticeship on an approved data or AI standard. Skills England maintains the standards, and several map onto AI-related roles. But they are not equally usable, and the gap between what's technically eligible and what you can actually run at scale is where we watch employer budgets go to waste. So here is the ranking, with the reasoning shown.

Level 4 is the route you can actually use today, at scale. At Level 4, the Data analyst standard (ST0118) carries a maximum government funding band of £15,000 (Skills England, 2021), and it is the vehicle for building applied data-and-AI capability in people who already work with data; we deliver it as Advanced Data & AI.

A Business Analyst route can deliver AI transformation capability, and the Data Protection & Information Governance Practitioner standard (ST0967, the same standard iO-Sphere delivers as our Data & AI Governance programme) is the funded home for AI governance work. There is no standalone "AI governance apprenticeship", so ST0967 is the honest, funded way in.

Level 4 carries no level-scoped age restriction (unlike Level 7, below), and its delivery market is deep: the analyst standard alone had 5,620 leavers in 2024-25 (DfE, March 2026). Crucially, it's the level at which applied capability forms for the people who carry the day-to-day work. If you're funding AI capability for teams that already touch data, this is where your money should go first.

Level 7 is where employers buy the wrong thing for the wrong reasons. The Artificial intelligence (AI) data specialist standard (ST0763) sits at Level 7 with a £17,000 band (Skills England, 2021). It looks like the prestige route, and that prestige is exactly the trap. From January 2026, Level 7 apprenticeships are only government-funded for apprentices who, at the start of training, are aged 16 to 21, or under 25 with an Education, Health and Care (EHC) plan or care experience (gov.uk apprenticeship funding rules 2026-27).

For a 22-or-over new start, which in practice is almost every senior AI hire an employer wants to fund this way, neither your levy funds nor government co-investment are available. You self-fund, in full.

Employers chasing L7 to badge their senior data staff are, in our view, usually solving a status question, not a capability one, and they discover the funding isn't there only after they've built the plan around it. If the outcome you need is applied AI capability across a team, L4 delivers it and L7 doesn't. L7 costs you cash on top.

Level 6 exists. A Machine Learning Engineer route sits at Level 6 (the ST1398 standard) with a £22,000 funding band (Skills England; bands are revised periodically).

Units (Level 5) are a hedge against a plan you haven't made. Skills England approved three Level 5 AI leadership units on 21 April 2026: AI strategy and opportunity (AU0009), AI adoption, procurement and governance (AU0010), and AI delivery and organisational transformation (AU0011). Each carries £750 maximum funding and is deliverable from 28 April 2026 (Skills England, AU0009). These are 30 to 140 delivery hours over 1 to 16 weeks, for employed learners aged 19+, and they confer no regulated qualification.

We don't deliver units, and you should know that before we say this: a unit is a funding innovation, not a training model. A few dozen hours ends before the practice hits the real edge cases, which is where AI capability actually forms.

The only employers for whom a unit is the right first spend are narrow: someone needing a first structured contact with the field, a team bridging to a fuller route already planned, or an employer who can fund nothing else this year. If you buy units because you haven't yet decided what capability you need, you've bought the hedge instead of the plan. Judge any levy spend the way you'd judge cash: by what the learner will have done by the end.

Which of these does iO-Sphere deliver, and when we're not your answer

iO-Sphere delivers Levels 3 to 5 only, and every programme in the line is built with AI integrated from the ground up, not bolted on:

Data & AI Essentials: Level 3, on the Data Technician standard (ST0795), the entry rung before L4
Advanced Data & AI: Level 4, on the Data Analyst standard (ST0118), this page's worked example
AI Transformation: Level 4, on the Business Analyst standard (ST0117), for teams putting AI to work
Data & AI Governance and Data & AI Strategy: Level 4, both on ST0967, the funded vehicle for AI governance capability
Data Engineering: Level 5, on the Data Engineer standard (ST1386)

Data is the substrate AI runs on, and data engineering is the plumbing all of it really runs on: an AI-capability plan that funds only the roles with AI in the job title skips the layer everything else stands on. All of these are funded through the levy for eligible employers, and for most AI-capability queries L4 is the right starting level, with L3 the rung for cohorts earlier in the journey.

Here is when iO-Sphere is not your answer, said plainly:

If your target learners already operate at Level 5 and up and need genuine L6/L7 depth, our Level 4 programmes are a foundation, not the destination. You should fund that destination directly through the relevant higher-level standard or a self-funded degree-level programme.
If you need data science or machine-learning engineering rather than data analysis or AI transformation, we don't deliver it. The honest route is the Level 6 Machine Learning Engineer standard or a specialist data-science programme, and that path will serve you better than forcing the need into a L4 shape.
If you need a regulated qualification at Level 5 or above, the Level 5 AI leadership units won't give you one, and neither will our L4 apprenticeships in the way a formal L6/L7 qualification would. Fund the qualification-bearing route directly.

We build Level 3 to 5 apprenticeships. Our cohorts are recent, so there is no achievement rate to show you yet: our numbers will appear in the same DfE tables as every other provider's when cohorts complete, and you should hold us to that rather than take our word for it. Where your need is above our ceiling, the responsible thing is to point you at the route that meets it, not to sell you the floor.

Step-by-step: accessing your levy funds for AI training

Funding is plumbing. The order that matters comes before it: decide the capability first, then fit the funding to it. Here's the sequence that works.

1. Map the capability before you look at the calendar

The question that opens the budget is which roles need to be doing what by next year. Ask it once a year, ideally with your CFO in the room.

The map has four standard uses, and most levy plans only ever consider one of them. Graduate intakes run as funded programmes, where the levy pays the training and you pay a salary you would have paid anyway. Retraining for people whose roles are changing, where the funded route keeps the institutional knowledge a redundancy-and-rehire cycle throws away. Data and AI capability for the teams that carry the P&L, which is where this page's routes live. And transfers to suppliers whose skills gaps slow your own delivery. That map is the innovation; the training calendar comes after.

2. Check the standard on Skills England

Confirm the standard you want is approved for delivery and note its current version and funding band on the Skills England standard page, never a provider's marketing page, which may quote a superseded figure. For a Level 4 data analyst route, that's the £15,000 band on ST0118.

3. Put five numbers on one page

This is the worksheet the whole argument comes down to. Five numbers, all published, none of which any provider's brochure will assemble for you:

#	The number	Where it comes from	For the AI routes on this page
1	Your monthly inflow	0.5% of pay bill, less the £15,000 allowance	Your payroll data
2	The clock on it	Funds contributed from 1 Aug 2026 expire after 12 months (down from 24), and the 10% top-up is gone (gov.uk funding rules 2026-27)	Your existing pot keeps its prior terms
3	Cost per head	The standard's funding band (Skills England; revised periodically)	ST0795 Data & AI Essentials £13,000 · ST0118 Advanced Data & AI £15,000 · ST0117 AI Transformation £18,000 · ST0967 Data & AI Governance £10,000 · ST1386 Data Engineering £19,000
4	Your cash on top	2026-27 rules v3	£0 for any apprentice aged 16 to 24 at every employer; for 25+: 5% non-levy, 25% at a levy payer with an exhausted pot
5	Working time	The standard's published off-the-job hours, inside paid working hours	ST0795 348 hours · ST0118 370 · ST0117 348 · ST0967 370, which we schedule as about six hours a week; ST1386 sets 418, scheduled across the programme

Row 5 is the one nobody volunteers, and it is the row the whole decision should turn on. The levy pays the band; your business pays the hours. A funded programme that doesn't change what someone can do wastes the dearer of the two, which is the arithmetic behind everything this page argues.

Three budget lines sit outside the levy but belong on the same sheet: levy funds never pay wages or travel; a £2,000 hiring payment exists for a non-levy employer taking on a 16-to-24 apprentice from 1 October 2026 (flagged as withdrawable in-year, so confirm before you plan on it); and employer National Insurance is 0% for apprentices under 25 earning up to £50,270.

4. Choose a provider and agree the programme

Brief a provider against your capability map, not a generic "AI course" request. Agree the standard, the funding band, and how off-the-job training will be timetabled: it must happen within the apprentice's normal, paid working hours. That's a legal protection, not an optional extra.

5. Set it running through the service account

Add the apprentice, confirm the funding, and the training draws down from your account. If your pot won't cover it (a levy payer whose funds run dry), you co-invest 25% from 1 August 2026 for apprentices aged 25 and over; an apprentice aged 16 to 24 at the start stays 100% government funded even then (2026-27 rules v3, amended 29 July 2026).

Using levy transfers to fund AI upskilling beyond your own pay bill

Transfers are the most underused line in the system: in our experience, almost no large employer we speak to has costed what its expiring funds could buy at a supplier. A levy-paying employer can transfer up to 50% of its unused annual funds to another employer (raised from 25% on 22 April 2024).

Run the calculation that changes the decision: any levy you will not spend before it expires is worth zero to you as cash, but it can fully fund AI capability at a supplier or partner whose skills gap is slowing your own delivery. You benefit from the capability without employing anyone new, and the alternative is watching the funds expire. The mental shift is small and specific: the levy is not a cost line, it is a transferable asset with a deadline.

On the receiving side: an employer with a pay bill under £3 million doesn't pay the levy, but is not shut out. Such an SME can fully fund apprenticeships two ways: through government co-investment, or by receiving a levy transfer that can cover up to 100% of the training cost. Not paying in is not the same as not being able to benefit. See levy transfer for the mechanics.

Choosing an AI training provider and programme that fits the funding rules

Judge a levy-funded programme the way you'd judge one you were paying cash for: by what learners will actually be doing, and who's coaching them. A funded course that doesn't change what someone can do is the most expensive training there is, because it spends the dearest resource in the building: working time.

Our view, and the reason we build the way we do: people get good at data and AI by doing the work, coached by people who've done the job, not by studying it in lectures. On the programmes that have it, our learners work in Prism, a simulated e-commerce company built on 500M+ rows of real data. Real data, simulated company; the sandbox is the safety.

When you shortlist providers, ask what learners produce, on what data, assessed how. The assessment method is set by each standard's assessment plan, and the common pattern on data and AI standards is a portfolio of real work plus a professional discussion rather than an exam hall, which matters if a written exam is what has stopped people applying.

Common pitfalls and eligibility gaps to check before you commit

Level 7 age eligibility. From January 2026, a 22-or-over new start on the Level 7 AI data specialist route gets no levy funding and no co-investment. This catches employers planning to fund senior AI staff. Check ages before you plan.
Fund expiry tightening. From 1 August 2026, contributions from that date expire after 12 months, not 24, and the 10% government top-up on new funds is removed. Panic spending near an expiry deadline is the levy wasted twice: shelf-ware consumes working time as well as budget.
Units aren't apprenticeships. The three Level 5 AI leadership units (£750 each) confer no regulated qualification and won't build sustained capability on their own. They're a bridge, not a destination.
Non-levy is a pay-bill test. "Non-levy" means a pay bill under £3 million. It is never a staff count.
Off-the-job training is protected time. It must sit within paid working hours. A provider who suggests learners "catch up at home" is offering something that breaks the funding rules.
The funded route may not be the right route. If the outcome you need isn't served by any funded standard, pay cash for the right thing rather than take the funded wrong thing.

What iO-Sphere believes about this

Capability, not technology, is the bottleneck for AI. The evidence is stark and consistent: 95% of enterprise generative-AI pilots delivered no measurable P&L return despite $30 to 40bn of investment, and the report named the core barrier as learning, not infrastructure (MIT NANDA, "The GenAI Divide: State of AI in Business 2025"). RAND found more than 80% of AI projects fail, twice the rate of non-AI IT projects, with the top root cause being misunderstanding the problem to be solved (RAND, report RRA2680-1, August 2024).

S&P Global Market Intelligence found the share of companies abandoning most of their AI initiatives jumped from 17% to 42% in a single year (Voice of the Enterprise: AI & Machine Learning Use Cases 2025, fieldwork Oct to Nov 2024).

Read together, dated and separate, these say the same thing: AI initiatives fail on adoption, the demand side, not on tools. That is not a rhetorical flourish; it changes which funded route you should choose. It is why we weight a coached Level 4 programme measured in months over a 30-hour unit: adoption capability forms in sustained real-work practice, not hours of structured contact, and a unit ends before the practice reaches the edge cases where capability actually sets.

It is also why the levy is best understood as a capability budget rather than a spend-rate target. If the training doesn't change what the learner does on Monday, the same adoption-failure rate that sank those pilots applies to your programme too. You'll have spent budget and working time to buy a badge, not a behaviour. Buy tools last, build capability first.

Next steps: building an AI-capable workforce with levy funding

Don't start with a training calendar. Start with a funding map: what capability each team needs over the next year or two, and where doing-based training compounds. Graduate intakes run as funded programmes. Retraining for people whose roles are changing. Data and AI capability for the teams that carry the P&L. Transfers to suppliers you depend on. Then spend against the plan.

If you're a levy payer weighing how to fund AI capability for your teams, our employer apprenticeships page sets out how our Level 4 AI Transformation programme funds through the Growth & Skills Levy: a five-figure programme that costs an eligible employer a small co-investment or, in the fully-funded cases, £0. If your need is above Level 5, we'll say so and point you to the route that meets it. Explore the routes, or talk to us about mapping your levy against the capability you actually need.

FAQ

Can you use the Growth & Skills Levy for AI training?

Yes. Levy funds in your apprenticeship service account can pay for approved data and AI apprenticeship standards: for example the Level 4 Data analyst (ST0118, £15,000 funding band) or a Business Analyst route delivering AI transformation. From April 2026 the levy also begins funding shorter, more flexible training. The levy is not apprenticeship-only money in the way many employers assume.

Which AI apprenticeship standards can the levy fund?

Skills England-approved data and AI standards, including the Level 4 Data analyst (ST0118, £15,000), the Data Protection & Information Governance Practitioner (ST0967, the funded route for AI governance), and higher-level routes such as the Level 6 Machine Learning Engineer and Level 7 AI data specialist (ST0763, £17,000). Level 7 is only government-funded for new starts aged 16 to 21, or under 25 with an EHC plan or care experience, from January 2026. This makes L4 the route most employers can actually use at scale.

How much is the Growth & Skills Levy and who pays it?

It's 0.5% of an employer's annual pay bill, paid by employers whose pay bill exceeds £3 million a year, collected monthly through PAYE. A £15,000 annual allowance offsets the charge. The rate, threshold and allowance are unchanged since the levy began in 2017: only the name and the range of fundable training have changed.

Can a small business that doesn't pay the levy still fund AI training?

Yes. A business with a pay bill under £3 million doesn't pay the levy but can still fully fund apprenticeships: through government co-investment, or by receiving a levy transfer of up to 50% of a larger employer's unused funds, which can cover up to 100% of the training cost. Not paying in is not the same as not benefiting.

Are the new Level 5 AI leadership units worth funding through the levy?

They fit narrow cases: a first structured contact with AI leadership, or a bridge while a fuller route is planned, and each is funded at £750. But at 30 to 140 hours they confer no regulated qualification and won't build sustained capability on their own. We don't deliver units, and we say so plainly: judge them by what a learner will have done by the end, not by the fact they're funded. If you're buying a unit because you haven't yet decided what capability you need, you've bought a hedge instead of a plan.

What changes for the levy from 1 August 2026?

Three mechanics tighten: the 10% government top-up on new funds is removed, funds contributed from that date expire after 12 months instead of 24, and a levy payer who exhausts their pot co-invests 25% for apprentices aged 25 and over. Full funding widens to cover every apprentice aged 16 to 24, at levy and non-levy employers alike, even where a levy pot is empty (amendment of 29 July 2026): £0 to the employer. Policy correct as of August 2026; check the latest DWP funding rules before you budget.

When is iO-Sphere not the right provider for AI training?

If your learners already work at Level 5 and up and need genuine L6/L7 depth, if you need data science or machine-learning engineering rather than data analysis or AI transformation, or if you need a regulated qualification at Level 5 or above, our Level 4 programmes are a foundation, not the destination. In those cases fund the higher-level standard or a self-funded degree-level route directly. We deliver Levels 3 to 5, and where your need sits above that ceiling we'll point you to the route that meets it.