Everything here is a working draft for the partnership to agree. The figures are built from days and rates rather than picked as headlines, and the approach section sets out the line we would take on every scored question.
Existing AI provision is built for businesses that can already specify what they need. They know the process they want automated and they procure against it. Those firms are a minority, and they are not the firms this Fund exists to reach. Barnsley’s small employers know they are being told to do something with AI and have no way of knowing where it would help them. Provision that requires a specification excludes them at the first step.
So our cycle begins with diagnosis, not delivery. Every business receives a structured discovery before anything is designed: how it quotes, how it schedules, where the hours actually go. The build follows from what that finds. This is why the systems get used, and it is the part conventional provision skips.
Youth AI provision has the same shape and the same blind spot. It teaches people to use a chatbot well, which produces school leavers who are AI aware and not one of whom is AI employable. Employers in Barnsley are not short of people who can write a prompt. They are short of people who can look at a business process and build something that fixes it.
So the young participants on this programme are not taught tools. They are taught implementation, on live systems inside real local businesses, alongside the people building them. By the end of a cohort a participant can sit with a business owner and work out where the hours are going, specify what a system needs to do, build and test it, and check its output before anyone relies on it. Those are the skills currently being hired for and not found.
That is the difference between a certificate and a career. A young person leaving this programme can point at a working system inside a named Barnsley employer and say they built it. They also have a route onwards, because the instructor certification we document as part of this project is open to them. The capability stays in the borough in the people, not only in the businesses, and that is the long-term benefit this Fund is asking about.
A cohort is a group of businesses going through the programme together rather than one after another. It is the unit every option below is counted in, so it is worth being precise about what one contains.
| When | What happens | Who leads |
|---|---|---|
| Weeks 1–6 | Discovery, one business per week. How the firm quotes, schedules and where the hours go | ForeShiloh |
| Week 3 | Shared workshop day. All six businesses in one room with their staff and the young participants recruited locally | Fresh Meet |
| Weeks 5–12 | Builds, three weeks each, staggered one week apart so they never run all at once. Young participants work alongside the build, which is where the implementation skills are learned | ForeShiloh |
| 3 and 6 months after | Follow-up: is it still being used, and is the business extending it | Both |
Two reasons for grouping businesses rather than taking them one at a time. The workshop only works with a group, because the shared room is where people learn from each other’s problems rather than only their own. And it is the only way the numbers work: six sequential one-at-a-time projects would take far longer and cost considerably more per business.
Five costed ways to deliver this. Every total is the sum of its lines, which is what question 27 is scored on. All five sit inside the fund’s £100,000 to £350,000 range and complete inside the 12 months.
Priced per cohort, not per day. The unit is a cohort and what it delivers, which is how this ought to read to an assessor and what we agreed between us. There is nothing in the fund documents that asks for day rates. The Budget Management Tool wants costs attributed to roles and activities, profiled monthly, with notes justifying each one, so the underlying days live in the notes column as the working rather than on the face of the bid.
Equal partners, equal share. Both organisations carry the same total, so the partner budget splits straight down the middle. Neither of us could do this alone: one side brings the AI build capability, the other brings the employer and youth pipeline, the workshop delivery and the track record that makes the application possible at all. The AI build engineer sits as its own line, because that role is a person the grant hires for the project rather than income to either partner.
| One cohort of six businesses | Scale | Cost |
|---|---|---|
| ForeShiloh: discovery across all six, build lead and handover | 13.5 days | £8,775.00 |
| AI build engineer: five of the six builds | 17.5 days | £8,312.50 |
| Fresh Meet: cohort lead, workshop mentors, employer and youth recruitment | 15.7 days | £10,205.00 |
| Workshop venue and refreshments | 2 days | £900.00 |
| Learner materials | 6 businesses | £510.00 |
| Travel across the borough | 6 businesses | £840.00 |
| Cohort price | 6 businesses | £29,542.50 |
| Per business | £4,923.75 |
| Across the whole project | Cost |
|---|---|
| ForeShiloh: method documentation and open toolkit, telemetry and measurement build, instructor certification route, technical support at 3 and 6 months | £28,600.00 |
| Fresh Meet: project management and DSIT reporting, monitoring and evaluation | £21,450.00 |
| Delivery-side software licences, 12 months | £3,120.00 |
| Project total | £53,170.00 |
| Option | Businesses | Cohorts | Cohort price | Each partner | Total ask | Per business |
|---|---|---|---|---|---|---|
| A | 16 | 4 of 4 | £30,887.50 | £83,200.00 | £190,857.60 | £11,928.60 |
| B | 20 | 5 of 4 | £30,530.00 | £96,850.00 | £222,285.60 | £11,114.28 |
| Clower risk | 24 | 4 of 6 | £29,900.00 | £63,700.00 | £186,591.60 | £7,774.65 |
| D | 25 | 5 of 5 | £34,367.50 | £101,725.00 | £243,008.10 | £9,720.32 |
| Eproposed | 30 | 5 of 6 | £29,542.50 | £72,475.00 | £216,953.10 | £7,231.77 |
One finding shapes all five. Business builds cannot run in parallel: six businesses building in the same three-week block is around 21 days of specialist work inside 3 weeks, which no team absorbs. Staggering the builds one week apart resolves it, and beyond roughly 20 businesses a second AI build engineer is needed. That is not a cost to apologise for. A method only one person can run is not a legacy for Barnsley, and funding a second engineer is the clearest evidence we have for the long-term impact question worth 15 marks.
It also explains why A and B look expensive per business. Without an engineer, one specialist has to do every build, so the same fixed project work is spread across far fewer firms. C and E carry their overheads properly.
Smallest commitment, easiest to evidence. Weakest on reach, which carries 15 marks, and by some distance the most expensive per business.
A single specialist at full stretch all year, and hard to argue on value for money at this cost per firm.
Eight more businesses than A for less money, because a second engineer builds under the established method. The lowest ask of the five and the lower-risk version of the proposed shape.
Included for completeness. The largest ask and the dearest cohort, five fewer businesses than E, and it leaves almost no room if anything slips.
Greatest reach, cheapest cohort, lowest cost per business, and four weeks in hand.
We would propose E, with C as the fallback. E is strongest on the criteria that actually carry marks: 30 Barnsley businesses for reach, the lowest cost per business for value for money, and an engineer trained inside the method as evidence the capability outlives the funding. It is also the cheapest cohort of the five at £29,542.50 for six businesses, which is £4,923.75 a business for a working system and the staff trained to run it.
Two things have to be true for E, and neither is ours alone to guarantee: the pipeline has to reach 30 businesses rather than 24, and an AI build engineer has to be available across the year. If either looks uncertain when we compare notes, C is the same delivery model at a smaller scale and it is also the lowest ask on the table.
The scoring framework is published and the weightings are not evenly spread, which decides where the effort goes.
| Question | Topic | Marks |
|---|---|---|
| Q20 | Innovation and new knowledge generated | 20 |
| Q21 | Reaching businesses and cohorts in Barnsley | 15 |
| Q22 | Long-term benefit to Barnsley | 15 |
| Q23 | Monitoring and evaluation | 15 |
| Q19 | Meeting the needs of the Fund | 10 |
| Q24 to Q28 | Management, experience, workforce, budget, value | 5 each |
Innovation alone is worth 20 marks, more than any other question, and it is the one place where a genuinely different method beats conventional provision. The top five questions carry 75 of the 100 marks. Everything else is hygiene, where complete and consistent beats eloquent.
The line we would take on each scored question, and what is still needed before it can be written properly. Full drafts sit behind each of these.
The biggest question on the paper. The starting point is the one above: conventional provision serves businesses that can already specify what they need, so it never reaches the firms that cannot. Beginning with diagnosis rather than delivery is the first departure, and everything else follows from it.
We advance beyond existing provision in three further ways. Voice is the access mechanism, so participants speak to the system rather than type to it, which removes the largest single barrier for the cohorts this Fund exists to reach. The workforce builds the system rather than receiving it, so training and implementation are the same activity. And the system checks its own output before anyone sees it, using independent verification agents, one running on a different underlying AI model because a different model carries different blind spots.
The genuine research uncertainty is stated plainly: does voice-first, build-based upskilling produce materially higher sustained adoption among workers without higher education qualifications, and does it persist once the provider leaves? The published evidence measures completion and satisfaction, not six-month use by underserved cohorts. Nobody knows the answer, which is what makes it fundable research rather than delivery.
Scored on how precisely the target group is quantified and how real the access routes are. We already have a good deal: Fresh Meet and Bravand are Barnsley Chamber of Commerce members, Tracey Johnson and Matthew Snowden at Tech SY have agreed to introduce businesses and confirmed no conflict, and Ben Hawley at Enterprising Barnsley has offered a meeting.
Our strongest structural argument. Unlike training, every business still owns a working system the day the funding ends, and staff trained to extend it, so the benefit compounds rather than decays. We quantify hours returned per business per week against a baseline taken at discovery, name the government priorities causally rather than decoratively, and use the Chamber route to evidence spillover into local supply chains.
We also treat the negative effects honestly, which most bids skip and the framework explicitly rewards: displacement anxiety, over-reliance on AI output, and businesses becoming dependent on something they cannot maintain. Each has a real mitigation, including that every business buys its own tool subscriptions directly, so there is no lock-in to us.
The most under-rated question on the paper. It carries the same weight as reach and long-term impact and is usually written as an afterthought. We have a real advantage here: because we build the system, usage is measured from logs rather than surveys. Almost no training provider can say that, and the framework marks data completeness directly.
The plan is a full KPI table with a method and a rationale per target, the DSIT question set administered at discovery and at cycle end with a named owner per cohort, surveys completed in the room on workshop day rather than emailed afterwards, and telemetry at three and six months. Attrition, withdrawal and self-report bias each get a stated mitigation.
The framing for the whole bid, and it rests on the specification problem above. Barnsley’s small employers do not have an awareness problem. They have been told to do something with AI and have no way of knowing where it would help them, which is precisely what conventional provision assumes they already know. Over each cycle a business receives a bespoke system built around how it actually works, and the build is the training. Leadership and frontline staff learn it together, which satisfies the two-level requirement by construction rather than by bolt-on.
Hygiene questions, where complete and consistent beats eloquent. Fresh Meet leads project management as accountable body, with work packages, named owners and a risk register that must correlate with the budget. On experience, Fresh Meet’s delivery record carries the answer and ForeShiloh contributes the AI evidence: a live client platform in production, the verification method with its record of caught defects, and the instructor certification route.
On value for money we benchmark against the market rate for bespoke AI implementation, which most bids omit, and we evidence leverage: each business buys its own AI subscriptions directly, roughly £100 to £200 a month, which is employer commitment the grant does not pay for.