AI for Manufacturing
SMEs
99% of UK manufacturers are SMEs, and most of them are running a multi-million-pound operation on spreadsheets, a twelve-year-old MES and one bloke who knows what Line 3 does on a Friday. We help you put AI on the bits that move the needle, not the bits a consultant will deck-out for you.
99%
OF UK MANUFACTURERS ARE SMES (MAKE UK, 2024)
£150bn
GDP UPSIDE BY 2035 (MAKE UK/SAGE, 2024)
28%
SME MANUFACTURERS USING AI (MAKE UK/AUTODESK, 2024)
Manufacturing's productivity gap isn't a labour problem anymore.
UK manufacturing has been ~10% less productive per hour worked than Germany for as long as the ONS has measured it. The Made Smarter Review (BEIS, 2017) said industrial digital tech could add £455bn to the sector over a decade. Most SMEs are still nowhere near that.
AI isn't the hard bit. The hard bit is that off-the-shelf "AI for manufacturing" is sold to plants with 5,000 staff and a data team. SMEs don't have a data team. You've got a works manager, a spreadsheet and a shift that starts at six.
The opening isn't more software. It's a small number of AI loops wired into the systems you've already got.
TODAY'S SME PLANT
- Quoting from a spreadsheet, by feel
- Planning whiteboard, redone Monday
- Maintenance run-to-failure
- QC by eye, on a sample
- Tribal knowledge in two people's heads
WITH THE FIRST AI LOOPS
- RFQ to quote in minutes, priced consistently
- Schedule reflows when an order, machine or shift changes
- Sensors flag the bearing before it fails
- Vision QC on every part, not every tenth
- SOPs and tribal knowhow in one place the AI can read
Where AI earns its keep in an SME plant.
Skip the hypeware. These five have a track record in real UK manufacturing SMEs, most of them written up in Made Smarter case studies. Pick one or two to start, not all five.
Predictive maintenance
Cheap sensors on the spindles, motors and compressors that cost you when they go. The model flags the change before the part actually fails. Beverston Engineering's Made Smarter case study is the one most people get pointed at first.
Vision-based QC
A camera and a model that catches the defects your QC team is now sampling for. Every part, every shift. Cuts scrap, cuts the warranty claims you only find out about three months later.
Scheduling and forecasting
Demand forecast from order history, stock and lead times. Schedule that reflows when reality changes, not when the planner gets in on Monday. The version of APS your ERP doesn't have.
RFQ and drawing extraction
Customer sends a PDF, a STEP file and a spec sheet at half four on a Friday. The loop reads them, costs the job against your routings, drafts the quote. Estimator approves or edits. Hours becomes minutes.
Shop-floor copilot
SOPs, drawings, change notices and 20 years of tribal knowledge in one place your team can ask in English. Setter walks up, asks how Line 3 ran the last batch. Gets the answer, not a folder.
Where we come in.
We're a UK AI engineering team. We build agentic systems for SMEs, and we've put loops live for engineering firms, builders, recruiters, financial advisers and bedroom manufacturers. The pattern's the same every time.
No twelve-month transformation, no PowerPoint. Audit the plant, pick the loop with the biggest payback, launch it, then iterate.
BOOK A SHOP-FLOOR AUDITShop-floor audit
A day or two on site. We walk the lines, sit with the planner, the estimator and the works manager. We map your ERP, MES, OEE board, spreadsheets and where the numbers don't agree. You get a written list of loops, scoped and priced.
Plant brain
SOPs, drawings, routings, supplier specs, tribal knowhow and the contents of half a dozen shared drives, captured into one place the AI can actually read. Modelled around ISA-95 so it talks to your ERP and MES instead of replacing them.
First loop, live
Sensors or data feeds, model, tools the AI can call, a human-in-the-loop gate for the bits that need a sign-off. You see it run on your numbers, your parts, your customers. Your team sees the payback before phase two is even quoted.
Govern, then add the next loop
ISO/IEC 42001 for the AI management bits, your ISO 9001 quality system left intact, ICO and UK GDPR handled for any personal data. Then we move to the next loop on the list. Maintenance, then scheduling, then RFQs. Small steps, not big bangs.
Other UK SMEs already did it.
Three published Made Smarter and UKRI case studies. All UK SMEs, all named, all public. Useful when the board asks who else has actually done this.
Beverston Engineering.
Precision machining SME in the North West. Made Smarter-backed data capture and predictive maintenance on the machining cells. Triggered new engineering hires and a multi-million-pound capex plan. Case study published by Made Smarter.
Goodflex Rubber Co.
Rubber hoses, mouldings and assemblies for automotive, off-highway, food and pharma. Took a £20k Made Smarter West Midlands grant, added their own capital and put in EFACS E/8 ERP with scheduling and MRP. Published SME digital-transformation case study.
Digital Spare Parts Supply Chain.
NBT Group with Senseye and Northumbria University, £227k Made Smarter Innovation Challenge. AI on inventory and predictive maintenance for the spare-parts supply chain. Published by UKRI.
Sources: Made Smarter Adoption, Made Smarter Innovation Challenge (UKRI), Make UK "UK Manufacturing: The Facts 2024", Make UK / Autodesk "Future Factories Powered by AI" (Sept 2024 survey, 151 companies), Make UK / Sage "Making it Smarter" (2024), ONS International Comparisons of Productivity, BEIS Made Smarter Review (Maier, 2017), BCG "Shaking Up the Factory Floor with Digital and AI" (2024). Adoption figures are survey numbers, not a census.
Most manufacturing AI projects don't land. Here's why ours do.
BCG's 2024 survey of ~1,800 manufacturing executives found 68% had started implementing AI. Only 16% hit the targets they'd set. The other 52% didn't get value out the other end.
The reasons are the same every time. The data wasn't there. The model worked in a slide deck and not on the shop floor. Nobody owned it. The works manager was never bought in. The thing that was meant to launch in eight weeks turned into an eighteen-month system-integration project.
We've spent the last three years engineering against exactly that list.
HOW WE STAY ON THE 16% SIDE
- One loop at a time. We pick the loop with the cleanest data and the clearest payback, and build that first. Not a digital twin of the whole plant.
- Human-in-the-loop, on day one. Estimator approves the quote. Planner approves the schedule. QC engineer reviews the flagged parts. The AI doesn't go autonomous until the humans trust it.
- Live on your numbers. Not a six-month pilot in a sandbox. You see the loop run on real orders, or we've done something wrong.
- Wired into ERP and MES, not replacing them. ISA-95 model, your existing systems stay. The AI sits next to them, not on top.
- You own the code. Source, models, data, IP. We can host it, or hand it to your team. No vendor lock-in, no rev-share, no per-seat licence creep.
The ones manufacturers ask first.
We're not a data-rich plant. We've got an ageing MES and Excel. Are we too far behind?
No. Most of the UK SMEs Made Smarter has worked with started exactly there. The first loop is usually the one that creates clean data as a byproduct, not the one that needs it first. We've helped engineering firms with paper job-cards, not a perfect digital backbone.
Is this Made Smarter? Can we get the grant?
We're not the Made Smarter Adoption programme. They run the matched-grant scheme through regional partners. If you qualify, get on it: it's good money and a proper diagnostic. We deliver the work alongside or after. Quite a few of our clients used Made Smarter to pay for part of the audit, then hired us for the build.
What about the EU AI Act, ISO 9001, the ICO?
If you sell into the EU, the AI Act applies to you. After the May 2026 Omnibus deal, the high-risk Annex III obligations are now due by 2 December 2027. The general Article 50 transparency rules still kick in on 2 August 2026, with the Article 50(2) watermarking obligations deferred to 2 December 2026. Your ISO 9001 quality system doesn't go anywhere. We sit AI alongside it, often within an ISO/IEC 42001 management wrapper. Anything that touches personal data we handle under UK GDPR with the ICO's guidance.
How do you handle health and safety on the shop floor?
AI doesn't replace HSE controls. Safety-critical interlocks stay in PLCs and hardware, the way the HSE expects. The AI sits above that, recommending and routing work, not overriding a safety system. Anything that touches a person physically gets a documented risk assessment before it goes live.
Does this rip out our ERP / MES?
No. We've yet to meet an SME that wanted to replace its ERP in the same year it adopted AI, and we wouldn't recommend it. We connect to Sage, SAP Business One, Microsoft Dynamics, Epicor, Infor, custom MESs and the spreadsheet that secretly is your MES. ISA-95 model in the middle so swapping any of those out later doesn't break the AI.
Our works manager isn't convinced. He's seen a lot of "transformation" projects.
Good. He's usually right. We won't run the project without him on board. The audit involves a couple of hours with him, the planner and the estimator. If after that any of them think the loop won't pay back, we say so and don't take the build. We'd rather not start than launch something the shop floor ignores.
How much does it cost?
Audit is fixed-fee. First loop is priced before we start, scoped against the audit. Ongoing loops fixed per phase. We tell you the number on the call. If a Made Smarter grant applies, we'll help you scope the application against it.
Bring us the bottleneck.
Tell us where it hurts. Quote turnaround, scrap rate, maintenance, planning, the bit that costs you every month and nobody's got time to fix. Thirty minutes, no slides. You'll get a clear read on whether AI is the right answer for that loop, and what it would take.