Background

Large organisations make decisions constantly, but rarely see their consequences clearly. A choice made in sales affects operations; a change in R&D timelines ripples into HR and finance; a procurement decision reshapes logistics months later. These interdependencies are understood intuitively by experienced staff but are almost never mapped explicitly — which means that when a decision goes wrong, the causal chain is reconstructed in hindsight rather than anticipated in advance. The result is a recurring pattern: decisions that looked reasonable in isolation produce outcomes that surprise everyone 12 to 18 months later. A simulation environment that makes these interdependencies visible and explorable before commitment could fundamentally change how organisations reason about consequential choices. The "fake company" framing — using synthetic data rather than live organisational data — removes political sensitivity and lets participants experiment freely with scenarios that would be too risky or too controversial to test in reality.

Problem

Key questions to be answered in the project:

1. How can the causal dependencies between organisational functions be identified and mapped reliably enough to serve as the foundation for a simulation?
2. What level of model fidelity is needed for a simulation to produce outputs that experienced staff recognise as plausible — and what simplifications are acceptable without losing credibility?
3. Which decision types benefit most from simulation — where is the gap between intuitive reasoning and modelled reasoning largest?
4. How should the simulation surface uncertainty and model limitations so that users develop calibrated trust rather than over-relying on its outputs?

In this project we aim to...

• Produce a validated dependency map of a fictional organisation by applying decision archaeology: identifying one recurring, high-stakes decision type and reconstructing its causal effects across functions through structured interviews and workshops.
• Build a simulation layer on top of the dependency map that allows users to run "what if" scenarios and observe how decisions in one function propagate through the organisation.
• Design and prototype the human-AI interaction model — how users pose scenarios, how results are presented, and how uncertainty is communicated.