A chain with a weakest link highlighted.
Making decisions in the face of uncertainty is complex and often scary. The toughest part of the challenge is figuring out which uncertainty is worth reducing, and at what cost.
On the one hand, you can often reduce your uncertainty if you’re willing to make an investment in learning. On the other, you have decisions that may or may not be influenced by the unknown.
To know which uncertainty is worth reducing, I like to think about how the quality of my decision will improve, as a function of that investment. If I can spend $10 or ten minutes to learn a little, what should I learn about?
I adapted a model from reliability engineering to help answer this question. Reliability engineering asks how you can best strengthen the overall chain. If your weakest link is very weak, you almost always want to strengthen that one. If you have lots of similar links, maybe you can improve all of them with one design change, and that will have the biggest impact.
Applying this to reducing uncertainty – but with a goal of better decisions – I picture a collection of unknowns. Each unknown has a learning curve showing what $1, or $10, or $100 will reveal.
Figure 1: The five unknowns have individual learning curves showing how an investment in resources will reduce the unknown. One example has no uncertainty reduction, while the other four vary.
If we had a budget to reduce the uncertainty, we might want to invest in reducing all of them, or we might have a single weakest link we need to prioritize. It depends on how the overall decision will change as a function of the uncertainty reduction.
Figure 2: The five sources of unknowns also will have different impacts on the best decision. As the uncertainty is reduced, we may find ourselves on a new leaf of the decision tree where the unknowns no longer matter.
To see how these separate sources of uncertainty could change the decision, we need to chain them together. This is where reliability engineering comes in as sometimes the risks are all in series, and you look at the decision that comes through the other end. This is the “weakest link” analogy.
Figure 3: The series of unknowns, which together will lead to an outcome based on our decisions and how much uncertainty we choose to learn. We would probably want to invest in the unknown with the steepest learning curve first, as long as it impacts the decision.
In other cases, some part of the uncertainty may set us down a different path. In reliability engineering Figure 4 shows a top path where unknowns 2 and 3 are said to be “in parallel” compared to unknowns 4 and 5. If this is how the unknowns and decisions are interconnected, it might be worth learning which path we’re on as a first priority.
Figure 4: There are unknowns in parallel, depending on unknown #1. We will end up caring about unknown 1, 2, and 3 OR 1, 4, and 5.
You now have a model for recognizing what you should try to learn so you can reduce the unknowns that will have the biggest impact on the decision, for the lowest investment.
