AI in Manufacturing: Practical Uses for UK SMEs, from Quotes to Predictive Maintenance

The trade press talks about smart factories and digital twins. The reality in most SME manufacturers is an inbox of RFQs, an ERP that half the business trusts, a maintenance whiteboard and a planner who keeps the schedule in their head. The useful question is which of those jobs AI can help with now, what data each one needs, and which to leave until the basics are in place.

This guide is for owner-led UK manufacturers of roughly 20 to 250 staff: the adoption data, seven uses in order of readiness, predictive maintenance in more depth, then the limits, the funding and a first project. If you are still deciding on the system that would hold your data, read off-the-shelf vs custom ERP for UK manufacturers first, because most of what follows depends on it.

How many UK manufacturers use AI

Fewer than the headlines suggest. In its analysis of the Business Insights and Conditions Survey for June 2026, the Office for National Statistics found around 22% of manufacturing businesses reporting use of at least one AI technology, mostly a single one. That compares with around 35% of all UK businesses with 10 or more employees and 58% in information and communication.¹

How manufacturers get their AI matters too. Among manufacturing businesses using AI or unsure whether they were, 49% were using free-to-use software, 35% bought software or ready-to-use services and 15% had developed something in-house.¹ Free tools are fine for trying ideas, but check their data terms first, as covered below.

Make UK's 2026 survey of its members gives the industry's own view. Just 2% of manufacturers said AI was widely embedded in their operations, and 18% had not adopted it at all; most of the rest were running pilots or small experiments. Where AI was in use, 83% applied it to business support such as HR, marketing, finance and administration, 24% to design and R&D, 11% to production, 7% to supply chain and logistics and 6% to quality control.²

The Department for Science, Innovation and Technology's AI Adoption Research, from 3,500 interviews in February to May 2025, suggests why the shop floor lags. It groups manufacturing with agriculture, mining and energy, so treat the figures as indicative. Firms in that group were more likely than average to have faced barriers implementing machine learning (51% against 28%) and computer vision (39% against 27%), and among those already using AI, 69% cited limited AI skills as holding back wider use.³

Seven practical uses, ranked by readiness

The ranking is our judgement, based on how often each task repeats, how easy the output is to check and how much data must exist first. Most firms can start the first three this quarter.

1. Quoting and RFQ handling

Much of an estimator's time goes on the parts of an RFQ that need no judgement: reading the drawing pack for material, finish, tolerances and quantities, finding the last similar job, and typing it all into a quote. AI is good at the reading and the finding. It can list the requirements, flag anything unusual such as a tight tolerance, and suggest past jobs with similar features.

Keep the cost model out of AI's hands. Machine rates, material prices, setup times and margins belong in a quoting tool or your ERP, where the numbers are fixed and auditable.

2. Reading supplier and customer documents and emails

A manufacturer's office spends hours a week moving facts between documents: customer purchase orders keyed into the ERP, acknowledgements checked against what was ordered, delivery dates updated from supplier emails, certificates of conformity filed against the right batch. AI does this well because each answer can be checked against the source next to it.

It works best with something to compare against: matching a supplier's acknowledgement to your purchase order and flagging a changed price or a later date. A person approves anything that changes an order or a price.

3. Sales and customer service assistants over your own product data

Customers and distributors ask the same questions every day: is this in stock, what is the lead time, will this part fit, where is my order. An assistant that answers from your own product, stock and order records, and says when it does not know, can handle the routine ones for your sales team or on your website. It only works if that data is in one place and current.

4. Production scheduling support

Planners carry a lot in their heads: which jobs run well together, which machine is temperamental, which customer will accept a split delivery. AI can propose a sequence, show the knock-on effect of a rush order or a breakdown, and flag jobs at risk of missing their date, but only with accurate routings, realistic cycle and setup times and a true picture of capacity. In many SMEs, getting those numbers into a system and applying simple scheduling rules does much of the work before any AI is involved.

5. Demand forecasting

For a make-to-stock manufacturer, a better forecast means less cash tied up in stock and fewer shortages. Models need several years of order history by product, with one-off orders and price changes marked. Set a simple baseline first, such as the same period last year adjusted for trend, and only adopt a model that beats it on your own data. For make-to-order firms, forecasting load on key machines or materials is usually more useful than forecasting products.

6. Quality inspection with vision

Camera inspection is one of the most talked-about uses and one of the hardest for an SME to get right. A vision model learns from examples, so it needs many images of good parts and of each defect you care about, under the lighting it will see on the line. A well-run process produces few defects, so collecting enough examples can take months, and reflective surfaces and changeovers all affect accuracy.

A realistic start is to fit a camera at one inspection point, record every part for a few months, and have inspectors label the defects they find. That builds the data set and shows whether the defects are visible to a camera at all. Make UK gives quality inspectors moving from manual checks to managing exceptions as an example of AI supporting workers rather than replacing them.² If a camera can see people as well as parts, the data protection rules in the limits section below apply.

Predictive maintenance: start with condition monitoring

Predictive maintenance means using data from a machine to predict when a part is likely to fail, so it can be repaired at a planned time rather than when it breaks. It sits at the end of a progression that most maintenance teams will recognise.

• Reactive: fix it when it breaks.
• Planned preventive: service it on a fixed interval of time or running hours, whether it needs it or not.
• Condition-based: measure something that shows wear, such as vibration or temperature, and act when it crosses a limit.
• Predictive: use the trend in those measurements, and the history of past failures, to estimate how long is left.

What sensors and data it needs

The common measurements are vibration on spindles, motors, pumps and fans; temperature on bearings, gearboxes and electrical panels; motor current, which changes as a tool or load wears; and oil condition on hydraulics. Newer machines may already record some of this. Older ones can take clamp-on wireless sensors.

The sensors are the easy part. The prediction needs history: readings from the same machine through several failures of the same kind, matched to a maintenance log that records what failed, when and why. Most SMEs do not have that, which is why the sensible first step is condition monitoring with alerts on simple limits. It is useful from the first alert and builds the history that prediction needs later.

When it pays

Condition monitoring pays where unplanned downtime is expensive: a bottleneck machine that holds up everything behind it, a long lead time on spare parts, or a failure that scraps work in progress. It pays less on machines you have spares for, or where failures give no measurable warning.

Whatever the technology, the legal duty does not change. The Provision and Use of Work Equipment Regulations require work equipment to be maintained in an efficient state, in efficient order and in good repair, any maintenance log to be kept up to date, and maintenance to be carried out safely by competent people.⁴ Monitoring helps you meet that duty, and its alerts and the actions taken belong in the same log.

The limits: data, integration, people and data protection

Data quality comes first

Every use above depends on complete, correct records: part numbers that mean one thing, routings that match how jobs run, a maintenance log that is filled in. Make UK found that firms with weaker digital foundations are significantly less likely to see returns from AI investment.²

Integration with your ERP

An AI tool that cannot read from and write to your ERP creates another place to copy data into. The quoting assistant needs your cost model and the document reader needs your open orders. Plan how each use connects before you start, and expect the connection to be most of the work. We build these connections as part of our custom software work, and the same question decides whether a packaged tool will fit.

People and skills

The barrier Make UK members report most is capability rather than technology. Over half of manufacturers identified skills and capability gaps as their main constraint on using AI effectively, and half said staff do not have time to train.² Involve the estimator, planner, inspector or maintenance engineer affected from the start. They know where the data is wrong.

Data protection

Machine data is not usually personal data, but emails, customer records, operator logins and camera footage often are. The Information Commissioner's Office says the data minimisation principle requires you to identify the minimum amount of personal data you need for your purpose and process only that, and that you must not collect personal data on the off-chance it might be useful in the future.⁵

Cameras on the line and machine data tied to named operators can amount to monitoring workers. The ICO's guidance says you must carry out a data protection impact assessment before any monitoring likely to cause high risk to workers, and that if you go ahead you must tell workers about it before you begin.⁶ The ICO notes that guidance is under review following the Data (Use and Access) Act, so check the current version. For free AI tools, find out whether the provider keeps or trains on what you send before anyone pastes in a customer drawing or a price list.

Funding: Made Smarter adoption support

Made Smarter runs an adoption programme for small and medium-sized manufacturers in England. It offers impartial technology advice, digital roadmapping workshops, leadership and skills support and, in several regions, match-funded grants of up to £20,000 for hardware and software, at the time of writing (October 2026). The technologies it supports include artificial intelligence and industrial internet of things sensors, which covers the condition monitoring described above.⁷

Support is open to manufacturing or engineering firms with 249 employees or fewer, a turnover under £36 million or a balance sheet under £18 million, and a significant part of their operations in a participating region. Grants cover up to 50% of the cost, and the exact offer varies by region.⁸ In the West Midlands the programme covers Greater Birmingham and Solihull, the Black Country, the Marches, Stoke-on-Trent and Staffordshire, and Worcestershire.⁷ If you are in the region, see our software development in Birmingham page.

Where to start

Pick the job that repeats most often, measure it, and run a small pilot with a person checking every output. For most SME manufacturers that is quoting or document handling, with condition monitoring on one critical machine as the first step on the shop floor.

Before you pilot anything, put a number on the manual work. The manual work cost calculator estimates what repetitive admin costs you a year, which gives you a baseline to judge the pilot against. When a task proves itself, the next step is an AI agent that does it inside your own systems, with permissions and checks built in rather than relying on someone copying and pasting into a chat window.

Better systems are half of growing a manufacturing business. The other half is winning the work they will handle, which we cover in marketing for manufacturers: what works in the UK.