AI is in your factory…
but it’s stuck in the office

Walk through most UK manufacturing businesses today and you’ll find two very different worlds. In the office, someone’s using Copilot to draft a customer email and ChatGPT to tidy up a quote. While out on the floor, production still runs on job sheets, spreadsheets, and the know-how of an operator who’s been there for about 20 years.

Today, research shows AI is predominantly used in back-office functions, with 83% using it in HR, finance and admin. In contrast, just 11% use it in production, and only 2% claim AI is widely embedded across their operations.

So, if AI is already in your factory, but it’s failed to make it past the office door, the big question is WHY? And the answer is very different in manufacturing than it is for an accountancy firm or a marketing agency.

Most AI advice is written for businesses where the work happens at a desk. Manufacturing is different:

A wrong answer costs more than a redraft

When AI gets something wrong in an office, you rewrite an email. When it gets something wrong near the production line, you’re looking at scrap, downtime, a customer complaint, or a safety incident. It’s why manufacturing leaders are more worried about AI accuracy than any other group. It’s also why your processes are validated, documented and change-controlled, so AI can’t simply be dropped in.

Much of your most valuable data belongs to someone else

Customer drawings and specifications shared under NDA. Supplier pricing. Tooling designs. A lot of what passes through a manufacturer’s office isn’t yours to share, and your customers trust you to protect it. You’re also a bigger target than you might think. Manufacturing has been the most attacked sector for five years running, accounting for 27.7% of all cyberattacks in 2025.

Your data lives in two worlds

Office data sits in your ERP, your inbox, and your shared drives. Production data sits in machines and control systems that were never designed to share it, as well as on paper job sheets and spreadsheets. The government’s AI Champion for Advanced Manufacturing names ageing equipment and scattered industrial data among the main reasons AI gets stuck in a “pilot trap”. And it’s why 47% of manufacturers lack confidence in the accuracy and completeness of their data.

Your expertise isn’t sitting at a desk

The people who know your processes best often work shifts, don’t have a company email address, and rarely sit in front of a screen. Many are also nearing retirement: over 70% of manufacturers now have employees working beyond retirement age. Meanwhile, 50% say their people don’t have time to train. It’s therefore no wonder Make UK concludes, “The biggest barrier to scaling AI is not technology, but capability.”

And here’s the twist…

The good news is manufacturers know how to control a process. You do it every day through your quality management system, health and safety procedures, and change control. AI needs that same discipline.

Map AI by data, not just by tool

Ask every team which AI tools they use and what for. Then sort what you find by the data involved; using AI to draft a job advert carries a very different risk to pasting a customer’s drawing into a free chatbot.

Draw a hard line around customer data

Check what your NDAs and customer contracts say about sharing information. Then make one rule impossible to misread, for example, “Customer drawings, specifications and pricing never go into a public AI tool.”

Treat AI like any other process change

If someone wants to use a new AI tool, run it through the same kind of approval you’d use for a new supplier or a process change. Question: what data goes in? Who checks the output? What happens if it’s wrong? Because the route is familiar, people are far more likely to follow it.

Choose an owner who spans the office and the floor

AI shouldn’t sit with IT alone. Assign ownership to someone who understands both worlds, like your operations director, and give them the time and authority to make decisions. Ownership makes a measurable difference. When leaders are accountable for AI, 57% strongly agree it delivers meaningful business value, compared with 21% where they’re not.

With control in place, a move onto the production floor becomes the natural next step. The trick is choosing back-office use cases that lead somewhere:

• Quoting and estimating today, means better production costing tomorrow.
• Summarising non-conformance reports today, allows you to spot quality trends on the line tomorrow.
• Organising maintenance logs today, delivers predictive maintenance on equipment tomorrow.
• Capturing your most experienced operators’ know-how in clear work instructions today, keeps that expertise in the business when they retire tomorrow.

Each step reuses the controls you’ve already built: the same data rules, the same approval route, the same owner. That’s how momentum builds, one proven step at a time.

It’s a journey we know well. For a global stairlift manufacturer, we connected its fleet of installed units, so it could move from reactive to proactive maintenance. Phase one delivered live monitoring and remote diagnostics. Phase two lays the groundwork for AI-driven predictive analytics.

We won’t pretend AI belongs everywhere on the factory floor, because some decisions should always rest with the people who’ve spent years learning the job. But where it does fit, the manufacturers that got their back office in order first will get there faster, and with far fewer surprises.