Decision-Making Under Deep Uncertainty: What to Do When There’s No Historical Data to Rely On

You make decisions without historical data by shifting from prediction to robustness: instead of forecasting one likely future, you test how a decision performs across a wide range of plausible futures and pick the option that holds up reasonably well across most of them. Deep uncertainty means you don’t just lack a clear answer. You lack the data to even assign reliable probabilities to the alternatives.

What Makes a Decision Fall Under “Deep Uncertainty”?

Most uncertainty in business planning still has some historical anchor. You might not know next quarter’s sales exactly, but you have years of past data to build a reasonable probability range around it. Deep uncertainty is different. It shows up when there’s no comparable past to draw from, such as entering a market that doesn’t exist yet, planning around a regulation that hasn’t been finalised, or preparing for a climate scenario with no historical precedent.

Standard tools still have a role once the range of outcomes is at least loosely defined. Monte Carlo simulation can test how a decision performs across that range once you have one to work with, though under deep uncertainty, defining that range is the harder problem.

A few signals suggest a decision falls into this category:

• No comparable historical data exists for the situation at hand
• Experts disagree on the range of plausible outcomes, not just the most likely one
• The timeframe involved is long enough that conditions could shift in ways nobody can currently predict
• A single wrong assumption early on could invalidate the entire analysis

When several of these apply at once, treating the situation like an ordinary forecasting problem tends to produce false confidence rather than a useful answer.

How Is This Different From Ordinary Risk Analysis?

Ordinary risk analysis still assumes you can assign reasonable probabilities to different outcomes based on data or informed judgment. Deep uncertainty breaks that assumption at the root. If you don’t know the range of plausible outcomes with any confidence, running a simulation across an assumed range just bakes a false sense of precision into the results.

That doesn’t make simulation the wrong tool. It means simulation answers a different question than the one deep uncertainty actually poses. It’s built to explore outcomes within a known range, not to tell you what that range should be in the first place.

What Is Robust Decision Making, and How Does It Handle This Gap?

Robust decision making is a formal approach built specifically for situations where assigning probabilities isn’t credible. Instead of asking “what’s most likely to happen,” it asks a different question: “which decision performs acceptably well across the widest range of plausible futures, even the ones we can’t currently rank by likelihood.”

RAND’s Pardee Center for Longer Range Global Policy developed and formalised much of this methodology, working through cases where traditional forecasting broke down entirely, including long-range infrastructure and climate planning where historical data simply couldn’t answer the question being asked.

The practical shift is significant. Instead of optimising for one predicted future, robust decision-making stress-tests each option against dozens or hundreds of different plausible futures and looks for the choice that doesn’t fail badly in any of them. It trades a small amount of upside in the “best case” scenario for protection against the scenarios nobody can currently rule out.

How Do You Actually Apply This to a Real Decision?

This process takes longer than a standard forecast-and-decide approach, and that extra time is the honest tradeoff. For decisions with high stakes and a long timeframe, that cost is usually worth paying. For smaller, reversible decisions, a simpler approach is often good enough.

FAQ

How do you make decisions without historical data?

You shift from forecasting a single likely outcome to testing how each option performs across a wide range of plausible futures, then choosing the option that holds up reasonably well across most of them, rather than the one that performs best in only one predicted scenario.

Is Monte Carlo simulation useless under deep uncertainty?

Not useless, but limited on its own. Simulation assumes you can define a reasonable range of inputs to sample from. Under deep uncertainty, that range itself is the unresolved question, so simulation works better once other methods have narrowed the plausible range.

What is robust decision making?

It’s a formal approach that tests decisions against many plausible futures instead of one forecast, then favours the option that performs acceptably across most of them. It was developed largely to handle long-range planning problems where historical data can’t support a reliable single prediction.

Does this approach apply outside of government and policy planning?

Yes. While it was formalised for large-scale policy and infrastructure decisions, the same logic applies to business decisions involving genuinely novel situations, such as new markets, unprecedented regulatory shifts, or long-term investments with no comparable historical case.

About Author

Talmas Lomla

See author's posts