How to Make Better Decisions Under Uncertainty

In early 2020, a CTO I know had to make a call: invest heavily in remote work infrastructure, or wait and see how the pandemic played out. The data was ambiguous. The timeline was unclear. Nobody could model what was coming.

She invested. Not because she had better information, but because she had a better framework for deciding without complete information.

Most of us freeze when we don't have enough data. We delay, we ask for more analysis, we schedule another meeting. But the uncomfortable truth is: the most important decisions in your career will always be made under uncertainty. The question isn't how to eliminate uncertainty. It's how to decide well despite it.

The Myth of Perfect Information

We've been trained -- by school, by engineering culture, by the scientific method -- to believe that better data leads to better decisions. And often it does. But there's a point of diminishing returns.

Jeff Bezos calls it the "70% rule": if you wait for 90% of the information you'd like, you're probably too slow. Most good decisions can and should be made with about 70% of the information.

The remaining 30% is where judgment, frameworks, and courage come in.

Framework 1: Reversible vs. Irreversible

The single most useful distinction in decision-making under uncertainty is whether the decision is reversible.

Reversible decisions (two-way doors): pick a database, choose a framework, hire a contractor, try a new workflow. If it doesn't work, you can change course. Decide fast. The cost of being wrong is low.

Irreversible decisions (one-way doors): sign a five-year lease, sell the company, delete production data, ship a breaking API change to thousands of customers. Go slow. Get more data. Stress-test your reasoning.

Most decisions are reversible. We treat too many of them as irreversible, which leads to analysis paralysis.

Framework 2: Weighted Scenarios

When the future is genuinely uncertain, don't try to predict it. Instead, imagine three scenarios and weight them:

• Base case (60%): The most likely outcome based on current trends
• Upside case (20%): Things go better than expected
• Downside case (20%): Things go worse than expected

For each scenario, ask: "What would I do?" If the same action makes sense in all three scenarios, it's robust. If it only works in the upside case, it's a gamble.

A real example: choosing between two job offers. Company A is stable, good salary. Company B is a startup, lower salary, meaningful equity.

• Base case: Company B grows steadily, equity is worth modest amount. Total comp similar to Company A.
• Upside case: Company B hits product-market fit, equity is worth 5x your salary gap.
• Downside case: Company B fails in 18 months. You lose the salary difference and have to job search.

Can you survive the downside? If yes, Company B might be the better risk-adjusted bet. If the salary difference means you can't pay rent, it's not.

Framework 3: Bayesian Updating

Start with a belief. Update it as new information arrives. This sounds obvious, but most people either:

• Stick with their initial assessment no matter what (anchoring bias)
• Completely flip their view based on one data point (overreaction)

The Bayesian approach is calibrated: each new piece of evidence should shift your confidence a proportional amount, not all or nothing.

In practice: "I'm 60% confident this architecture will scale. After load testing, I'll revise that up or down based on results." Then actually revise it, rather than cherry-picking results that confirm your prior belief.

Framework 4: Expected Value with a Floor

Calculate the expected value, but set a floor on acceptable outcomes.

Expected value = (probability of success x value of success) + (probability of failure x cost of failure)

A bet with positive expected value is generally worth taking. But if the cost of failure is catastrophic (bankruptcy, job loss, reputational ruin), the expected value calculation doesn't matter. You need to survive to play again.

This is why professional poker players manage bankroll separately from individual hand decisions. The expected value of each hand matters, but never risking your whole stack matters more.

Putting It Into Practice

For a decision I'm facing right now, I'll walk through my actual process:

1. Classify: Is this reversible? (If yes, decide in 24 hours.)
2. Scenario plan: What are the three plausible outcomes? What would I do in each?
3. Check the floor: Can I survive the worst case?
4. Decide and record: Write down my reasoning in my decision journal.
5. Set review triggers: What new information would make me reconsider?

This process is informed by years of reading how the best capital allocators think under uncertainty. If you're looking for a structured collection of these decision scenarios and frameworks, KeepRule's scenarios section organizes them in a way that's practical for everyday use.

The Meta-Lesson

The goal isn't certainty. It's being comfortable with uncertainty while still acting decisively. The best decision-makers I've worked with share this trait: they don't know more than everyone else. They're just better at acting on what they know.

Start making faster decisions on reversible things. Invest your analytical energy in the irreversible ones. And keep a record of how your predictions compare to reality.

That's the whole game.