Skip to content
All articles
Industry 8 min read· by Lars Zimmermann

AI sparring: what managing directors really want to discuss

My first AI sparring session. The expectation was efficiency, the topic turned out to be incoming orders. What really occupied two managing directors.

In short

My first AI sparring session was hardly about efficiency. The bottleneck was incoming orders, not processing them. Two further topics appear in no AI brochure: distrust of the company’s own master data, and what happens when the one employee with the scheduling in his head is absent. If you do not know these three points, you talk past the company.

Auf Deutsch lesen: deutsche Fassung

I had planned an hour, had a self-built AI ERP in my luggage and a fairly clear idea of what we would talk about. Efficiency. Lead times. Where in the company the minutes get lost. After twenty minutes it was clear: frankly, that interested my two counterparts least of all. They had a different problem, and it was bigger.

The expectation was efficiency. The topic was incoming orders.

The sentence that turned the session came from the managing director, in essence: how do I use AI to win new customers? Right now that matters more to me than whether someone here works five minutes faster.

There I sat with my nice demo and had to swallow first. Because he is right. A company whose shop floor is half empty gains nothing when order entry gets faster. Efficiency pays off when there is work. If the work is missing, you are optimising idle time.

This is the point at which most AI offerings talk past the company. They come with process optimisation into a room in which revenue is being discussed. Both are legitimate, but the order decides whether anyone listens to you.

On top of this there is market pressure that you find in no AI presentation. Several of this company’s customers are relocating to Eastern Europe on group instructions, at prices that nobody here can or wants to quote against. Via manufacturing portals, enquiries come in that are well above the later award price and in the end still go abroad. Anyone in this position understandably only half listens when they hear “process optimisation”.

My stance on this is unchanged: I do not do dumping prices, and I no longer bend down for them either. The way out of the price war does not lead via the last cent, but via things a cheap supplier does not deliver with the part: reliable dates, clean documentation, fast and robust quotations. That is exactly where technology can contribute something.

What I have made of this: I now ask first where the bottleneck is. Not where the process is stuck, but where the business is stuck. Sometimes it is the shop floor. Often enough it is sales, visibility, the quality of quotations or the speed at which a quotation goes out. And yes, there are sensible AI applications for this corner. They just have different names from the ones everybody shows.

“In theory yes, in practice no”

The second sentence I took away is the most honest objection to automation I know. In essence: in theory I agree with you. In practice I do not. Our stock levels are not right. Now and then someone cuts a piece off and does not book it.

That is not resistance to technology. That is professional experience. Anyone who knows that their master data is full of holes does not trust any automation that builds on this data. And they are right.

My answer to that is uncomfortable, but it hit home: the quantities are not right today either. It is just that nobody notices, because a person sits in between who balances it out in their head. Automation does not make this error bigger. It makes it visible.

That is why the order is not negotiable: first the process, then the tool, then the AI. Anyone who automates a bad procedure gets bad results faster and in greater numbers. The clean process is not a precondition you talk yourself into. It is the work itself.

And one more thing belongs to honesty: at the start such a system never runs without errors. It learns from real cases from your company, not from a brochure. Anyone who keeps quiet about that is selling a promise they cannot keep.

The question after which the room briefly went quiet

At some point we came to scheduling. Booking bills of materials, allocating material, estimating dates. That is done by hand there, from experience, by feel. One man does it. He does it well, and has done for years.

Then I asked, provocatively: what happens if he is absent?

I have asked this question in every sparring session since, because it reliably touches a nerve. Not because the answer is new, but because it is rarely said out loud. Parental leave, a longer illness, a move, at some point retirement. In many companies a central workflow depends on exactly one head.

Incidentally, that is not an AI topic but a governance topic. Knowledge that exists only in one head is not documented, not auditable, not transferable. As an auditor I see this constantly, and it is always the same pattern: it works superbly until the one time it does not.

AI can actually help here, but differently from what people think. Not as a replacement for the colleague, but as a tool that makes his approach visible. Writing down rules he applies anyway. Logging decisions he takes anyway. At the end there is a workflow that someone else can run as well.

The first step costs nothing but attention: record a meeting in which this colleague explains why he decides the way he does, and write down from it the rules he has long been applying. Not as a check, but as a safeguard. Most people are happy to do this if you explain why.

  • Which recurring decision is taken in your company by exactly one person, without it being written anywhere how they take it?
  • What would happen in the week in which this person is not there?
  • Who could step in, and how long would they need to get up to speed?

What I showed, and what really had an effect

I had the automated order entry from my own company with me. PDF in, customer configuration, drawing filed, prices and quantities taken over. Previously twelve minutes of manual work, now twenty-six seconds. That is my own company, my own figures, not a reference customer.

The demo went down well. But something else had the effect.

Spontaneously I called up the two men’s website and looked at what legal information is placed there. A few points had visibly not been touched for years. That is not legal advice, and it was not a demonstration either. It was what it is: a look from outside that nobody takes in everyday business.

This moment moved more than any slide. Because it was concrete, comprehensible in two minutes, and because it shows how much is left undone unnoticed in day-to-day business. Not out of negligence. Out of being busy.

It was important to me to name the difference between demo and operation openly. What I show runs in my company, with my data, after two years of work on it. It is not a product you install on Monday and that knows the company on Tuesday. Each of these automations first had to learn how we work, and a person signed off every rule along the way.

The second thing that stuck was a figure from my purchasing. I saved forty per cent there in one year. Not through AI, but because we stopped making two or three more parts than ordered, just to be safe. That is the uncomfortable truth behind many digitalisation projects: the biggest lever is often in a habit, not in a piece of software.

Why the format is called sparring and not consulting

I went into this conversation without an offer. No pitch, no presentation with a price list at the end. That was a deliberate decision, and it was the right one.

A company that is thinking seriously about AI for the first time does not need a salesman. It needs someone who has the same problems and says honestly what worked and what did not. I run a precision engineering company myself. I have the same price discussions, the same delivery times, the same question of where the next order comes from.

Either you move with the times, or the times move on without you.

A football coach once said that sentence to me, and it fits here better than any market forecast. It is not a threat. It is an invitation to start calmly, instead of acting out of fear or doing nothing at all out of defiance.

In the end the two wanted to sleep on it and announced a return visit to see the system live in production. For me that is the most honest result such a conversation can have. Not a deal, but interest in the reality behind it.

What you can take away from this hour

If you are thinking about where AI should start in your company, then not with the question of which tool everybody is currently recommending. But with three sober questions.

  • Where is my bottleneck really? In processing, or in incoming orders? The answer decides which kind of AI application makes sense at all.
  • Which of my data do I not trust? Everything that builds on it will only work once the process behind it is clean.
  • Which workflow depends on exactly one head? That is at the same time your biggest risk and your best first use case.

None of these questions needs technology to be answered. They need an hour of honesty. After that you know whether AI is your topic or whether something else comes first. Either is a good result.

One last point, because it almost always comes up in such conversations: the vague fear of doing something wrong. What am I allowed to do at all, what is prohibited? This uncertainty paralyses more companies than any technical hurdle. It can be resolved by going through your own use cases once, soberly, and classifying them. A tool does not resolve it. For the vast majority of things a mid-sized company intends to do with AI, the answer is pleasingly unspectacular.

What really counts, by contrast: before a system takes a decision that affects someone, it must be clear who supervises it and who answers for it. That is not bureaucracy, it is the same question you ask yourself with a new machine. The liability stays in the company, not with the provider and not with the software.

And if you notice that the questions are uncomfortable: that is normal. Uncomfortable questions are the ones that change something. The comfortable ones you can have any vendor ask you.

Share: LinkedIn E-Mail

Frequently asked questions

What is an AI sparring session?+

An open conversation about your own company, without an offer at the end. I show what works with AI in my own company and what does not, and together we look at where the bottleneck in your company really is. No pitch, no presentation with a price list.

Where should a mid-sized company start with AI?+

With the question of the bottleneck, not with the question of the tool. If the problem is in processing, automation helps. If it is in incoming orders, visibility, quotation speed and costing are the more sensible starting points. First the process, then the tool, then the AI.

My master data is unreliable. Is automation still worthwhile?+

The data is not right today either, it is just that a person balances out the gaps in their head. Automation does not make this error bigger, it makes it visible. That is exactly why the clean process is not a pretext for postponing, but the first step of the work.

What does the bus factor have to do with AI?+

Knowledge that exists only in one head is not documented, not auditable and not transferable. AI can help to make this approach visible: recording rules, logging decisions. The aim is not a replacement for the colleague, but a workflow that someone else can run as well.

Does AI pay off when capacity utilisation is missing?+

Efficiency pays off when there is work. If the work is missing, you are optimising idle time. Then the first look belongs to everything that brings new orders, and only after that to the lead time on the shop floor.

Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)

Last updated: 31 July 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.

Further reading

Questions about your own case?

In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.

Continue reading