Monte Carlo simulation.

Where Monte Carlo simulation fits on a long engagement.

Simulation is the right tool when the question is about a distribution rather than an average. Project completion dates, capital at risk, capacity under uncertain demand and insurance exposure are all cases where the tail matters more than the centre, and a single expected value actively misleads.

The output is only as good as the input distributions, and those are usually the weakest part. Assumed normality where the real data has heavy tails, or independence between inputs that actually move together, produces a distribution that is precise and wrong — and the precision makes it more persuasive than it should be.

What an assigned team does with Monte Carlo simulation.

Dependence between inputs is where most simulations go wrong. Modelling correlated risks as independent understates tail risk severely, which is exactly the region the simulation was built to examine.

Getting that right is specialist statistical work rather than an implementation detail, assigned under managed data services alongside the engineering that productionises it.

What we use Monte Carlo simulation for.

• Ranges for planning decisions A distribution of completion dates or costs, rather than a single figure nobody believes.
• Tail risk quantified Extreme outcomes estimated where the average is not the decision-relevant number.
• Dependence modelled honestly Correlated inputs handled as such, because independence understates the tail.

How Monte Carlo simulation capacity is assigned.

Simulation capacity is assigned inside data science work, with input distributions documented as assumptions rather than embedded silently in code.