For whoever takes the case to the person who signs off spend: each cost line named, each value line backed by dated national evidence, and a one-page case to adapt.
Data & AI Apprenticeship ROI: Building the Business Case
By James Cotton · Last updated · 7 min read
Part of our topic guides on AI Skills for Business and Government-Funded Data & AI Training.
By James Cotton, Founder, iO-Sphere
You have a role in mind, perhaps someone already on the team, and a budget holder who will ask what it returns. Every number below is national and dated. We have no completion or retention figures of our own, so the evidence is the Department for Education's, with the year of each figure given where it is used.
The ledger a budget holder will recognise
The budget holder reads a proposal as a ledger, so give them one, with the source you would cite beside each line. The cost figures are for one route, Advanced Data & AI.
| Line | Side | What goes against it | Where the number comes from |
|---|---|---|---|
| Salary | Cost | What the role is worth to you, paid throughout | Your own pay scale |
| Off-the-job training | Cost | At least 370 hours, about ten working weeks | Skills England standard record, ST0118 |
| Manager review time | Cost | Hours a week spent setting and checking real work | Your manager's own estimate |
| Co-investment | Cost | Set by the apprentice's age and your levy position | Skills England funding band; funding rules 2026-27 paragraphs 213 to 214 |
| Productivity | Value | Work done that would otherwise queue or go outside, growing as the programme runs | DfE Apprenticeship Evaluation 2023, employer survey |
| Organisation-specific skills | Value | Capability built on your own data and systems | DfE Apprenticeship Evaluation 2023, employer survey |
| Retention | Value | Two years of capable work after the assessment | DfE Apprenticeship Evaluation 2023, both surveys |
Salary, hours and manager time come before the fee
Salary comes first. It is the largest line, and it exists whether or not the person trains. An apprentice is employed in a real role and paid for it, so price it at what the role is worth to you and remember that it keeps running through the training hours.
The protected time is next. On Advanced Data & AI the off-the-job minimum is 370 hours, about ten working weeks at 37.5 hours, spread across a programme whose length the apprenticeships overview gives route by route. Those hours are paid and unavailable to the day job. Cost them at salary.
Third is the line manager, the line easiest to leave blank. Someone has to set real work, review it and say what is wrong with it, most weeks, for the whole programme. Training is the provider's job. Reviewing the apprentice's own output stays with you, and it belongs in the ledger as hours from the manager's week.
Last and smallest is the co-investment: a share of the £15,000 Advanced Data & AI funding band that falls to nothing for a start aged 16 to 24, and for an older start depends on your levy position, £750 at a non-levy employer and £3,750 at a levy payer whose funds are spent. Check the employer National Insurance on a younger apprentice's pay before the salary line is final.
The value lines, and the evidence behind each
The Department for Education's Apprenticeship Evaluation 2023 employer survey put its questions to 3,926 employers (IFF Research, fieldwork 27 July to 27 October 2023, published 27 September 2024). 77% reported increased productivity. Where ICT was the main subject, 80% did. And 84% said they would recommend an apprenticeship.
Productivity goes first on the value side because the budget holder can test it against the team's backlog. Name the reports, checks and analyses that wait, or go to an agency. Early on the apprentice takes the routine reporting and data checks, later the analysis nobody had time for, a sequence the data analyst apprentice employer guide follows across the first year. Price each piece at what it would cost to buy in.
Skills relevant to the organisation were reported by 86% of the employers in that survey, and by 91% where the apprenticeship was at Level 4 or above. For data roles this line is concrete. A person trained on your definitions, your systems and your untidy tables brings that knowledge to every later piece of work.
Retention decides the size of the return, and it is where the 2023 figures fell furthest from 2021. Of employers whose apprentices completed at Level 4 to 7, 77% had kept at least one and 64% had kept all of them, against 89% and 81% two years before. Model it with your own turnover alongside the national figure.
Why the window is the two years after
During training the apprentice costs the most and produces the least unaided. Their work is supervised, reviewed and sometimes redone. The return arrives once the same person does that work alone and starts checking other people's. That happens after the programme, so the case has to run past it.
Apprentices were surveyed in 2023 as well: 4,919 of them, questioned between 4 July and 15 September that year. Of the completers, 95% were in work. 68% were still with the same organisation, and among those six to 18 months out the figure was 72%.
Asked about the job itself, 90% of those completers said they were better at it, 48% had been promoted and 75% had had a pay rise since completing. Put the pay rise in the ledger. It is part of what keeping a capable person costs, and the promotions show where their work tends to go next.
The crossover point is the month the apprentice's work stops needing a second pair of eyes before it ships. It varies by role and by manager. Ask the line manager to name the month they expect it, write it into the case, and count capable work from there to two years past the assessment.
Sensitivity: what breaks the return
The case rests on three conditions. If any one fails, the value lines shrink however good the funding looks.
- A role with nothing to apply the training to. The productivity and skills figures above come from apprentices working on live data; strip that out and the protected hours buy a qualification and nothing the business can use. The productivity and skills evidence above comes from apprentices doing the work, so check the role before you check the funding.
- No review time from the manager. If nobody reviews the work, the supervised period stretches and the crossover slides out of the two-year window. Budget the hours before the start, in the manager's diary.
- Non-completion. Figures the Department for Education released on 16 July 2026 put the overall achievement rate for 2024/25 at 65.4%, up 4.9 points, with retention up 4.8 points. That is a national figure across all apprenticeships, and it means roughly a third of the apprenticeships counted did not end in achievement.
Enter non-completion as a probability. The cost at risk is the salary and hours spent before a withdrawal, and the value lost is the whole two-year window. Weighting the case by that rate, rather than assuming completion, is what makes it survive a sceptical reading.
The one-page case to hand over
Fill the brackets from the lines above. Send it as the body of an email a day before the meeting, so the budget holder has read the numbers before anyone argues about them.
We propose to train [an existing colleague / a new hire] as an Advanced Data & AI apprentice in [team]. The costs are salary at [amount], at least 370 hours of protected training, [hours] a week of [manager]'s review time, and a co-investment of [amount for the start's age].
We expect [named work] during the programme and unsupervised work on [named problems] for two years after. National evidence from the Department for Education's 2023 apprenticeship evaluation supports our productivity and retention assumptions.
The risks are non-completion, manager time and [role risk]. We will measure the return by [measure].
Frequently asked questions
Is there a figure for the return per pound?
Not one an employer can use directly. The "£26 to £28 per £1" sometimes quoted is a 2015 Department for Business, Innovation and Skills net present value to the economy per pound of government funding, for Level 2 and 3 starts in 2013/14, which some providers present as employer ROI. A September 2021 report that Cebr wrote for St Martin's Group, a membership body, with co-funding from City & Guilds and NCFE, found an early employer net benefit of £2,496 during training for 2020/21 starters, in a year with COVID-era grants.
What if they leave after qualifying?
Then the value lines stop at the month they go, and the case should show a shortened window. When an apprentice did leave, 83% of the employers in the 2023 evaluation said the decision was the apprentice's. The levers are the ones that keep anyone: a real role at the end, a manager who uses their work, and a pay conversation at completion.
How many hours will the manager need, and how do we know?
The evidence is the manager's own diary. Take how long they spend reviewing a colleague's report now, multiply by the pieces the apprentice will produce in a week, and add time to set the work. Revisit the estimate at the crossover month, when review should begin to fall away.
Does this apply to an existing employee?
Yes. For a colleague already in post the salary line is already on the books, so the new costs are the protected hours, the manager's time and any co-investment by age. The role still needs real data work for the value lines to hold, and what they already know about the business is part of what you keep.
Will it fit your cohort?
Tell us your cohort size and what your people need to be doing by next year, and we will say whether we are the right shape, including when we are not.