I think of it like a person playing poker. Poker is often a great analogy for life because it involves making decision with incomplete information. Given the particulars of a situation you can use probability math to calculate the expected outcome. Should act in the manner that maximizes your expected outcome. Even still, for any given situation there's a chance you'll make the correct decision (based on the math) and still get the negative outcome.
Yet outcomes are the bottom line and without improving them, your efforts don't amount to much. My mind tried to reconcile these apparent conflict. Then I remembered a video I saw a while back. It talks about a measurement model for software development.
It starts with Outcomes and then traces back through Decisions, Information, and finally Metrics (ODIM for short). The point the presentation makes is that since Outcomes are a lagging indicator (they have already happened) they are useful for verifying behaviors but not managing them. For that you need a leading indicator (which for him is Decisions) that in theory will produce the desired Outcomes. That theory can then be testing by observing if you get the desired Outcomes. Decisions require the right Information and to get that Information you measure the Metrics.
Here's the presentation (starts around the half-way point where he starts discussing what I summarized above):