Scope and detail
Set the boundary and the level of detail from the decision, not from the data available.
- Define the question first
- Set the model boundary
- Stop at the detail that changes the answer

A model is worth what its checking is worth.
Building a model is the straightforward part. Establishing that it reproduces the operation you already have, and saying where it does not, is what makes its answers usable.
A model nobody has checked against the plant still produces confident numbers. Explore five things validation has to settle, using the cards or arrows below.
Scope and detail
Input data
Baseline comparison
Assumptions
Stated limits
Set the boundary and the level of detail from the decision, not from the data available.
Check the data the model will rest on before the model rests on it.
Reproduce the operation that already exists before changing anything in it.
Write down what was assumed, and what the answer depends on.
Say plainly which questions the model can answer and which it cannot.

Explore all five challenges · Select a card or use the arrows
Validation is agreed before a model is built, not claimed after it. What it produces is a stated basis for trusting the answers, and a stated boundary on them.
A boundary and level of detail chosen from the decision rather than from whatever data happened to exist.
Data coverage, period and exceptions examined before the model is built on them.
The current operation reproduced and compared on agreed measures, with differences explained.
What was assumed, how much the answer depends on it, and who agreed it.
The questions the model supports, and the ones it does not, stated before anyone relies on it.
Validated model + Baseline comparison report + Assumption and limitation register
We agree what the model has to reproduce, and how closely, before it is built. The result is a model with a documented basis for trusting it and an explicit statement of where that trust stops.
Explore our simulation project supportThe boundary and the level of detail follow from the question being asked. Modeling finer than the decision requires adds build time, more inputs to be wrong about, and no additional accuracy in the answer that matters. We agree where the model stops at the start.
We look at what the data covers, which period it represents, and whether exceptional days are mixed in with normal ones. An input accepted without examination becomes an assumption nobody recorded, and it is usually the cause when a model and reality disagree later.
Before anything is changed, the model runs the current operation and is compared with it on measures agreed beforehand, typically throughput, utilization and queueing or inventory behavior. Where it differs, we explain why rather than adjusting until the numbers match.
What counts as close enough is agreed before the comparison is run. Deciding afterwards invites the standard to move to wherever the model landed, which is how a model comes to be described as validated without the word meaning anything.
Every assumption is written down, and the ones the answer is sensitive to are tested across a plausible range. A conclusion that holds only for one value of an uncertain input is reported as exactly that.
We state the conditions the model is valid for, the questions outside its scope, and the point at which it should be revalidated. A model used three changes past its basis produces confident answers about an operation that no longer exists.
With expert simulation consultation.
Tell us the decision the model has to support and what you can compare it against today. We’ll set out what building and validating it should involve.