AI FOR MANUFACTURING

What AI can do today in an industrial SME (and what is hype)

No vendor marketing and no data-science jargon. This is what actually pays back in a plant of 10 to 100 people, written by someone who has spent 10 years on factory floors.

Everything written about AI and industry falls into one of two buckets: the vendor selling their platform as if it were magic, or the technician explaining models as if you had a data department. Neither answers the question an SME owner asks: what do I do with this on Monday, how much does it cost and when do I get it back?

This section answers exactly that. Case by case, with numbers, and also saying where NOT to go.

What DOES work today

1. Copilots for plant paperwork

Procedures, work instructions, control plans, audit reports, 8D complaint answers. Tools that already exist (ChatGPT, Copilot, Claude), cost 0-25€/month, real savings of 3-6 hours a week per technical-office person. The best effort/return ratio TODAY.

2. Machine vision for quality

Cameras that catch defects in line. Mature technology, suppliers from ~5,000€ per inspection point. Worth it when escaped defects are expensive (claims, rework) and the defect is visible. Not worth it to "look for the sake of looking".

3. Predictive maintenance (with caveats)

Sensors + models that warn before the breakdown. Works on critical, expensive equipment (compressors, furnaces, continuous lines). In an SME with 15 varied machines it rarely pays: a well-run preventive plan in Excel gives you 80% of the benefit for 2% of the cost. Maintenance plan first, sensors later.

4. Assisted demand forecasting

Models that beat the planner's gut feel using your sales history. Useful with 50+ active part numbers and seasonality. If you make to order with 10 references, you don't need AI: you need a good capacity calculation.

What is hype (for now)

  • "The autonomous factory": it doesn't exist even in the multinationals advertising it. What exists are highly automated cells with plenty of people behind them.
  • "Digital twins" for SMEs: they need a level of sensors and data that costs more than the benefit. A pretty name for a six-figure project.
  • "AI that optimises your production by itself": any system promising to optimise without anyone measuring the process first is optimising on thin air. Measure first (OEE, costs, capacity), optimise later.
  • Agents that "run your plant": in 2026 they are still demos. In the office they're starting to be useful; on the floor, nowhere near.

The hype pattern is always the same: it promises the end result without asking you for the groundwork. AI amplifies what you already have. If your factory doesn't measure, AI amplifies the mess.

Where to start with 0 €

  1. This week: use a copilot to draft ONE plant procedure you've been putting off. Measure how long it took before and now. That's your first proof, free.
  2. This month: get your basic data in order - hourly cost per line, OEE of the bottleneck, downtime hours. Without this, no plant AI project makes sense later.
  3. This quarter: if technical paperwork eats your hours, formalise copilot use (company accounts, which data is NEVER pasted, who reviews). If escaped defects eat your margin, get 2 machine-vision quotes and compare with your annual cost of claims.

The articles in this section

Predictive maintenance in an SME: when it pays and when it's money down the drain Machine vision for quality: the honest buyer's guide Copilots in the technical office: ground rules so you don't leak your data

The base before the AI

Every plant AI project starts by measuring what isn't measured today. The e2b templates - line costing, OEE and capacity, maintenance plan - are that first step, and they cost less than one hour of any vendor's consulting.

Start with maintenance