In theory yes, in practice no: the most common objection to AI

When I suggest automating a workflow or bringing in an AI tool at a company, almost the same sentence comes back every time: in theory yes, in practice no, every job is different here. I know that sentence well. I have said it myself. And in the vast majority of cases, the problem is not the technology.

What lies behind the objection

The objection is rarely hostility towards technology. It is a genuine feeling: every job seems unique. In manufacturing that is often even true, no workpiece is quite like the next. But the process around it rarely is. Quote, order creation, material planning, invoicing, all of this runs on the same patterns every day, even when the parts differ. What gets confused is the individual product with the always-identical path to producing it.

The volumes are actually there today too

Anyone who says everything here is bespoke is still doing it today anyway. Just in their head, by hand, from scratch every time. It feels flexible, but it is expensive and error-prone. The same employee makes the same decision twenty times a day, only invisibly. And precisely where something is always slightly different, there is usually a rule that simply nobody has written down.

Bespoke rarely means there is no rule. It usually means the rule sits inside someone's head rather than on paper.

The right order: process first, then tool, then AI

The most common mistake is to start with the tool, or straight with AI. That way you automate the chaos, and faster chaos is still chaos. The order that works for me is a different one:

• Process: write down what actually happens, step by step. Where do rules apply, where are there genuine exceptions?
• Tool: only once the workflow is clear do you look for the tool that maps exactly that workflow, not the other way round.
• AI: it comes last, and only where patterns are involved that a person cannot take in quickly enough, for instance deriving the right proposal from thousands of past quotes.

AI is not a substitute for a missing process. It is an amplifier. Amplify order and things get better. Amplify chaos and things get worse.

How I approach it in practice

I do not look for the most exciting problem, but the most boring one with the highest frequency. Something that happens every day and always causes the same aggravation. Then I make the hidden rule visible. Often an afternoon with the people who do it daily is enough: when do you deviate, and why? What do you check in your head before you click? Only then do I decide whether a clearer workflow is enough, a simple tool will do, or AI is genuinely worth it. Sometimes the best automation is not software at all, but a clearer agreement.

Why this is also a governance question

Anyone who first describes a workflow cleanly before putting AI on top of it is, as a side effect, doing exactly what an AI management system under ISO/IEC 42001 requires: knowing what you do, naming who is responsible, understanding the risks. Order in the process is the foundation on which AI becomes accountable in the first place. Skip that step and you automate a risk that at least a person previously had in view. In theory yes, in practice no is therefore almost never a technology problem. It is a signal that the process is not yet clear. Make it clear, and in practice no quickly turns into in practice yes. Where that starts for you is something we can look at together in a free initial call.