Factory Physics - 10 decisions where your intuition sabotages you

After all, in a complex factory, intuition alone is not enough. Our brains are designed for linear relationships, not for non-linear queueing effects, feedback loops and variability. Experience does not solve this: precisely because these relationships are non-linear, we do not learn from them automatically.

You will recognise yourself in the situations below. That’s the point. ;-)

1. “The bottleneck must never stand still.”

Reflex: That machine costs X million; downtime is pure waste. Full = good.

Reality: A bottleneck operating at an average of 95% capacity creates queues that can barely clear. The slightest disruption ripples through the entire system.

At 70% utilisation, a brief 5-minute disruption may have little or no effect on the rest of the line. At 95% utilisation, that same disruption becomes a traffic jam that only ceases to be noticeable in the rest of the line after hours – sometimes an entire shift.

Logic: Due to the u/(1−u) factor, 95% capacity becomes a throughput time time bomb. A bottleneck needs protective capacity: a little spare capacity to absorb variability. Typically, the optimal capacity lies around 80–85%. 

2. ‘We must keep WIP as low as possible; that is lean.’

Reflex: A lot of WIP is bad. So: reducing WIP = always good.

Reality: Below critical WIP, the bottleneck runs dry and throughput drops. Lead time may even increase: you produce more slowly, orders wait longer. Too little WIP leads to starvation, frantic rescheduling and extra changeover times.

Logic: You want to aim for controlled WIP, not ‘as little as possible’. WIP below W₀ = starvation, WIP well above W₀ = suffocation. 

3. ‘We’ll put a few extra people on it.’

Reflex: More people → finished sooner.

Reality: If the constraint is not labour but machine capacity or flow variability, adding extra people mainly results in: more work in progress (WIP up), more consultation, more handovers, more micro-variation. The bottleneck remains a bottleneck, only now with more chaos surrounding it.

Micro-case: An assembly company added 4 operators to a line where machine capacity was the constraint. Result after 2 months: 15% more WIP, 8% longer lead time, same throughput. The extra hands pushed more work into the system, but the bottleneck couldn’t go any faster.

Logic: Extra people only make sense if they relieve the real bottleneck. Otherwise, you merely exacerbate arrival variation.

4. ‘We need to plan more tightly and in greater detail.’

Reflex: Things are going wrong because the planning isn’t detailed enough.

Reality: In a physically unworkable system, any plan will break down, no matter how detailed. You’ll just end up with more re-planning, not more stability.

Micro-case: A food company switched from weekly planning to hourly planning. The planning team grew from 2 to 5 FTEs. Result: average delivery reliability remained at 78%. What did increase: the number of re-plans per week (from 12 to 47) and frustration on the shop floor.

Logic: First, the system must be sound (buffers, WIP, variability); only then can planning become ‘fine-grained’. Otherwise, you’re building an Excel masterpiece on top of physical nonsense. 

↳ Aha moment: Planning is information processing, not variability reduction. You can’t plan chaos away – you can only tackle it structurally.

5. ‘Our OEE is top-notch, so we’re doing well.’

Reflex: Machines are running at 85–90% OEE. That’s world-class.

Reality: OEE says something about local efficiency, not about overall flow. You can have a machine with 95% OEE that mainly produces useless WIP that nobody wants. That WIP clogs up aisles, lengthens lead times and makes planning more complex. And this is assuming that the OEE figures haven’t been measured incorrectly in the first place.

Logic: The real danger: green OEE dashboards give a false sense of control. The question is not: “Is this machine running well?” The question is: “Does this machine help to shorten lead times and reliably deliver what the customer wants?” Note: OEE and utilisation are not the same thing. Changeovers are an OEE loss, but may be necessary to produce what the customer wants.

6. “Larger batches mean we lose less to changeovers.”

Reflex: Fewer setups = more runtime = higher efficiency.

Reality: Larger batches mean longer waiting times for other products, greater peaks in WIP/inventory downstream and much greater arrival variability. You might gain 5% on OEE, but lose 30% on lead time.

Logic: Batch size is a trade-off between changeover loss and variability in flow. SMED reduces the pain of changeovers in this trade-off; blind ‘batching’ worsens the flow. Moreover, the efficiency gain follows an asymptotic curve: beyond a certain batch size, there is hardly any further gain by increasing the batch size even further.

7. “We’ll keep some extra stock, then we’ll be safe.”

Reflex: More stock = extra safety.

Reality: Uncontrolled stock clogs your flow, lengthens lead times, hinders flexibility and masks structural problems. Extra stock only helps if it is in the right place (decoupling points), in the right form (raw material, WIP, finished product), and quantitatively aligned with your variability.

Logic: Simply adding stock is like randomly installing airbags in a car: high costs, little extra safety. Buffers must be deliberately placed. 

8. ‘We still have room to manoeuvre: our capacity utilisation is only 70%.’

Reflex: 70% is far from 100%, so there is still plenty of room for volume.

Reality: That 70% is an average. During peaks, capacity utilisation shoots much higher. Typically: 20% of the time >95%, 60% of the time ~70%, 20% of the time <50%. It is the periods of high capacity utilisation that, due to the VUT relationship, generate a disproportionate amount of waiting time.

Logic: Average utilisation tells us little. You need to look at how often you enter the red zone, not just at the average. 

9. “We’re at full capacity; we need an extra line.”

Reflex: Investing = growth = problem solved.

Reality: Investing can certainly create extra capacity, but this comes at a cost and can generate additional complexity and work. The crucial question is: is that excess capacity even necessary? Often, it is not the machine that is the bottleneck, but the way in which WIP, variability and planning are organised. If you invest without first optimising, you are buying expensive overcapacity whilst the real constraints remain. And if you do invest: which other buffers (stock, lead time) can you then reduce to make the investment pay off?

Logic: The correct sequence: (1) Identify the real bottleneck using data. (2) Address that bottleneck (WIP limitation, sequencing, variability reduction). (3) Allow all other processes, decisions and KPIs to adapt fully to the pace and needs of the bottleneck. (4) Only if you are still constrained do you increase capacity. In the majority of factories, you don’t even get to step 4.

Micro-case: A metal processor invested €2.3 million in an additional milling machine. After installation, it turned out that the real bottleneck was the shared measuring room – where each part passed through twice for inspection. Within six months, the new machine was idle 40% of the time, waiting for clearance.

10. ‘Lead time variation is simply typical of our sector.’

Reflex: Our sector is volatile; this is just part of it.

Reality: As soon as you start measuring WIP, utilisation, variability and buffers, you see that 60–80% of the chaotic lead time variation is self-inflicted: expediting, ad-hoc priority changes, incorrect batch sizes, misplaced buffers, push behaviour.

Logic: Factory Physics distinguishes between exogenous variability (market, weather, customer behaviour) and endogenous variability (your own decisions). The latter can be significantly reduced. “That’s just the way it is” usually means “we’ve never systematically measured and addressed it”.

Conclusion

This is not about being smart or stupid. It is about the fact that production systems do not behave as our intuition expects.

These errors are systemically logical. They arise because production systems are rarely designed with these laws of nature in mind — not because managers are incompetent.

Factory Physics is the lens through which you suddenly see why these 10 reflexes go wrong, what decisions you ‘should’ be making, and how you can substantiate this scientifically. 

In Article 4, we show what actually happens when you start managing your factory as a physical system.