People have been talking about AI in manufacturing for years. Smart factories, predictive maintenance, automated sales — it all sounds promising. But in reality, very little has actually changed inside most companies.
Here’s why it’s been so difficult:
The data is a mess.
Product data lives in PDFs, spreadsheets, ERPs, emails, and most often, in people’s heads. Organizing it is the first step — and where most projects already get stuck.
Products are complex.
It’s rarely just one product. There are hundreds of variants, accessories, and configuration rules that differ by project or region — many of them undocumented. Even experienced engineers struggle.
Generic AI doesn’t understand the domain.
ChatGPT can write emails, but it doesn’t know your products or how your business operates. Without being trained on real cases, it’s more confusing than helpful.
What about Copilot?
Looks promising on paper — but real implementation often costs between €500K and €1M. You’ll need months to prep data, rework workflows, and bring in high-salary specialists. Even then, the outcome is uncertain.
Old AI projects were too slow and expensive.
Big consulting firms, long timelines, custom tools nobody used. Many companies got burned, and just stopped trying.
And most internal AI initiatives fail for a simpler reason.
They start without a real problem to solve — just curiosity. No clear goal, no expected outcome, and no roadmap. Most fade after a prototype or a few meetings.
But now the situation is changing:
– More companies already have digitized product data
– AI can be trained directly on real documents and cases
– You don’t need to rebuild your IT systems to get started
– And results can come in weeks, not months
In the next few years, AI in manufacturing will finally become real — not as huge, complex systems, but as small, smart tools solving practical problems: answering questions, speeding up decisions, helping people choose the right product.
It’s finally happening. Quietly — but for real this time.
Here’s why it’s been so difficult:
The data is a mess.
Product data lives in PDFs, spreadsheets, ERPs, emails, and most often, in people’s heads. Organizing it is the first step — and where most projects already get stuck.
Products are complex.
It’s rarely just one product. There are hundreds of variants, accessories, and configuration rules that differ by project or region — many of them undocumented. Even experienced engineers struggle.
Generic AI doesn’t understand the domain.
ChatGPT can write emails, but it doesn’t know your products or how your business operates. Without being trained on real cases, it’s more confusing than helpful.
What about Copilot?
Looks promising on paper — but real implementation often costs between €500K and €1M. You’ll need months to prep data, rework workflows, and bring in high-salary specialists. Even then, the outcome is uncertain.
Old AI projects were too slow and expensive.
Big consulting firms, long timelines, custom tools nobody used. Many companies got burned, and just stopped trying.
And most internal AI initiatives fail for a simpler reason.
They start without a real problem to solve — just curiosity. No clear goal, no expected outcome, and no roadmap. Most fade after a prototype or a few meetings.
But now the situation is changing:
– More companies already have digitized product data
– AI can be trained directly on real documents and cases
– You don’t need to rebuild your IT systems to get started
– And results can come in weeks, not months
In the next few years, AI in manufacturing will finally become real — not as huge, complex systems, but as small, smart tools solving practical problems: answering questions, speeding up decisions, helping people choose the right product.
It’s finally happening. Quietly — but for real this time.
