What can AI actually do for a precision manufacturing business?

In a subcontract shop, the honest answer is that it works the front desk, not the machines. It answers the phone and the inbox, captures RFQs properly, gives customers an order status, and chases the paperwork that holds up despatch. It does not set a tolerance, choose a process, price a job or touch a control. The value is that your estimator and your engineers stop doing admin, not that a machine gets cleverer.

That distinction matters, because most of what gets sold as "AI for manufacturing" is aimed at the shop floor, needs a lot of setup to work, and pays back far too slowly for a business turning over a few million.

Start with the jobs that are actually costing you

The front office in a small manufacturer is doing five jobs badly at once, usually while also running the shop:

• Taking enquiries on the phone, by email, and through the website.
• Making sure an RFQ is complete enough to quote.
• Telling customers where their order is.
• Finding and sending certificates of conformity and material certs.
• Answering the same five questions: lead time, capacity, can you machine this material, are you ISO or AS9100, what is your minimum batch.

Every one of those is repeatable, evidence-based work. None of it needs engineering judgement. All of it currently interrupts someone who does.

What it looks like in practice

One agent, working four channels, drawing on one shared knowledge base of what your firm actually does:

• Voice. Answers every call, including out of hours, takes the enquiry properly, recognises existing customers from their number, checks before it promises anything.
• Email. Reads the inbox by what each message is actually about, not just the order it arrived in: RFQs and drawings to estimating, purchase orders and call-offs acknowledged and flagged to planning, cert requests handled, complaints routed to quality.
• Chat. Answers capability, material, tolerance and lead-time questions on your site, and turns a serious enquiry into a captured RFQ.
• Social. Catches LinkedIn messages from buyers and OEM procurement and routes the real ones into the same flow.

The shared knowledge base is what makes the answers specific to you. It holds your capabilities and machine list, the materials you work, your tolerances and inspection equipment, your accreditations, your lead times and batch ranges. That is why the answers sound like your firm rather than a generic chatbot.

What it will not do

Being specific about the limits is the fastest way to judge whether it is real:

• It does not read a drawing and set a tolerance.
• It does not price a job or commit a delivery date it has not checked.
• It does not decide whether to take work.
• It does not go near a machine, a control or a CAM system.
• When it is unsure, it says so and hands to a person rather than guessing.

Your estimator still quotes. Your quality manager still signs off. It clears the path to them.

Why now, and why the scepticism is fair

Owners in this sector have watched technology promises come and go, and they are right to be careful. Their world runs on tolerance and traceability, so something that confidently makes things up is worse than useless.

The numbers say the caution is widespread rather than unusual. Make UK, surveying 151 manufacturers with Autodesk, found only a third (36%) using AI in their manufacturing operations, and only 16% regarding themselves as knowledgeable about it. Across the wider economy the ONS puts business AI use at 25%, up 15 percentage points since it first asked in late 2023. So most of the sector is not behind, it is deliberately waiting to see something that works.

The right test is not "is AI impressive". It is this:

1. Does it stay in its lane, front office only.
2. Is it answering from our information, not the internet.
3. Can we see every conversation, and switch it off.

If a supplier cannot answer those three plainly, it is not ready for a shop floor business.