
Oct 8
/
Benjamin Schumann
Interview: Simulating Reality with AnyLogic
In August 2025, Dr. Oskar Schneider (Senior Data Scientist at Horváth and kind of an "influencer" in the OR world) interviewed me on my career in simulation modeling and supply chain applications.
This is a copy of the original interview.
This is a copy of the original interview.
Who are you?
My name is Ben and I haven’t worked for the past 15+ years. I preferred to build and play with toy worlds in my computer. Luckily, this also solves real-world issues for clients so it’s been win-win 😊
These toy worlds are simulations of reality. I am quite good at recreating factories, warehouses and supply-chains in a tool called AnyLogic. These become valuable if your real problem is caused by causal chains (“butterfly effect”), randomness and just things changing over time.
While Excel models are good for accounting and mathematical models are good for optimisation, simulation models are good for supporting real-world “nastiness”. I can include whatever tiny detail impacts operations…
We’ve all seen it during Covid or with the infamous Suez-canal ship: most supply-chain operations are already quite optimised, but not robust. In manufacturing, it is often reversed: factories will churn out no matter what but are far from optimal.
With simulations, I can support both: we build a toy version of reality and test a lot of “what-if” questions. Once we have a solid model, we can also optimise for whatever we like, but crucially including uncertainty, randomness and causality.
These toy worlds are simulations of reality. I am quite good at recreating factories, warehouses and supply-chains in a tool called AnyLogic. These become valuable if your real problem is caused by causal chains (“butterfly effect”), randomness and just things changing over time.
While Excel models are good for accounting and mathematical models are good for optimisation, simulation models are good for supporting real-world “nastiness”. I can include whatever tiny detail impacts operations…
We’ve all seen it during Covid or with the infamous Suez-canal ship: most supply-chain operations are already quite optimised, but not robust. In manufacturing, it is often reversed: factories will churn out no matter what but are far from optimal.
With simulations, I can support both: we build a toy version of reality and test a lot of “what-if” questions. Once we have a solid model, we can also optimise for whatever we like, but crucially including uncertainty, randomness and causality.
If you would need to model the supply chain of Amazon, how would you start?
I wouldn’t. Telling me “I need you to model the supply-chain of Amazon” calls for an immediate and loud “WHY?” 😀
If the client says they want to be able to do “stuff” with it, it is time to run away. Or educate the client about how models need to be purpose-driven: WHY do you want to model something?
Once the WHY is sufficiently clear, I paint the system like a painter. In broad strokes first, adding details only if and where necessary, iteratively.
This means that Jeff Bezoz would get a working model if his global supply-chain after 1-2 days. But it would be quite a rough sketch...🙂
I created an entire only course around this theme and called it “The art of AnyLogic modeling”.
If the client says they want to be able to do “stuff” with it, it is time to run away. Or educate the client about how models need to be purpose-driven: WHY do you want to model something?
Once the WHY is sufficiently clear, I paint the system like a painter. In broad strokes first, adding details only if and where necessary, iteratively.
This means that Jeff Bezoz would get a working model if his global supply-chain after 1-2 days. But it would be quite a rough sketch...🙂
I created an entire only course around this theme and called it “The art of AnyLogic modeling”.
How do you view the interplay between mathematical optimisation and simulation?
Optimisation is useful for problems that can withstand abstracting away time, causality and randomness. This works well for (some) scheduling, some SC-planning, etc.
However, it breaks down very quickly for the questions my clients have: How much will we produce next quarter? Will our optimal work plan for next week play out if Amanda is off sick?
Another difference I see: Simulations are far from black boxes like an optimization model.
Any client can follow along when I get my (virtual) Lego blocks out and start building their system virtually. They see the system animated. With “real” machines, vehicles, routes… And then: we play. Like I have been for the past 15 years 🙂
However, it breaks down very quickly for the questions my clients have: How much will we produce next quarter? Will our optimal work plan for next week play out if Amanda is off sick?
Another difference I see: Simulations are far from black boxes like an optimization model.
Any client can follow along when I get my (virtual) Lego blocks out and start building their system virtually. They see the system animated. With “real” machines, vehicles, routes… And then: we play. Like I have been for the past 15 years 🙂