Uncertainty is everywhere. Demand fluctuates, suppliers run late, machines break down, and the weather refuses to follow the forecast. Every decision we make is influenced by things that are not known in advance.
This matters significantly when we try to optimise. A good optimization model squeezes the most out of the resources available, which means the best solution often sits right at the boundary of what is feasible. It is most cost efficient if you have just enough capacity to meet all demand. But what if demand turns out to be slightly higher than you planned for?
Most optimisation models still exclude uncertainty, which is a shame. In this blog, we explain the three most common approaches to include uncertainty: robust optimization, risk constraints, and recourse actions. Each of these three methods makes a different trade-off between safety, cost, and flexibility.
Robust: be ready for anything
In a robust approach, you look for a solution that is feasible in every considered scenario. Whatever happens, your plan can be executed, which is the main advantage of this approach. The drawback is that robust solutions can be very pessimistic and therefore costly.
Consider a situation in which you need to invest in some capacity to meet demand. By protecting yourself against the worst case, you may end up with a lot of unused capacity in most situations.
Risk: accept a chance to fail
An alternative is to accept that you cannot protect against everything. Instead of demanding feasibility in all scenarios, you require it in a specified percentage of them, for example 95%. In the remaining scenarios, it is accepted that demand cannot be met.
This typically leads to much cheaper and less conservative solutions than the robust approach, because you do no longer need to cover the extreme cases. A drawback is that those scenarios are simply discarded during decision making. The model does not care how bad the outcome is in the 5% of cases it ignores. A scenario where you miss demand by a little is treated the same as one where you miss it by a lot.
Recourse: fix it when it happens
The third option is to build corrective actions into the model. Rather than ignoring the scenarios you cannot satisfy, you plan how to repair them once they occur. These are called recourse actions.
Recourse actions are usually expensive, but they give you a way out of a bad scenario. Think of placing an emergency order to meet unexpected demand or renting extra equipment to increase your capacity. You hope not to need them, but they provide an escape if needed.
Because the cost of these actions is part of the model, the price of fixing a bad scenario can be compared against the price of preparing for it in advance. This makes recourse a more balanced way of handling uncertainty, but it requires that effective corrective actions exist and that you can estimate what they cost.
Which approach should you choose?
There is no single best approach to include uncertainty in a decision-making process. The most appropriate method depends on the context. If failing to meet a constraint is very costly or dangerous, for example in healthcare or critical infrastructure, a robust approach is worth considering despite its conservatism. If there is a fallback option available, then considering recourse actions might be the best approach. In case violating a scenario is sometimes acceptable, a risk-constrained approach might be a better fit.
Uncertainty is always present, and ignoring it does not make it go away. Thinking about what happens when reality is different from what is expected is a good starting point to improve your decision-making process. How to implement the methods discussed will be covered in upcoming blogs, but if you cannot wait, feel free to reach out to us on info@doingthemath.nl.


