Decision Making Under Uncertainty: Test Your Options
7 min read
Decision making under uncertainty means choosing an action when outcomes, their probabilities, or both remain unclear. Test each option against plausible conditions, then compare its worst outcome and maximum regret. Regret is the gap between an option’s payoff and the best available payoff under the same condition.
Key Takeaways
1Remove options that violate non-negotiable constraints before comparing their potential returns.
2Use maximin to select the best worst-case payoff among the conditions you test.
3Use minimax regret to limit the largest payoff gap between your choice and the best alternative.
4Gather evidence that could change your choice, and record the conditions that would trigger reconsideration.
What Changes in Decision Making Under Uncertainty?
The main change is how much confidence you can place in probabilities. In Frank Knight’s classic distinction between risk and uncertainty, risk involves measurable probabilities, while uncertainty does not. Business decisions often combine both, so label your assumptions rather than forcing everything into one category.
Separate missing knowledge from inherent variability. The research paper Decision-Making Under Uncertainty: Beyond Probabilities distinguishes epistemic uncertainty, arising from limited knowledge, from aleatoric uncertainty, arising from randomness. A contract review can resolve uncertainty about cancellation rights. It cannot remove variation in future customer demand.
Keep the broader decision process simple. Use the six practical steps of structured decision making to establish your objective and options, then concentrate your uncertainty analysis on the assumptions that could change the preferred path.
Choose a Decision Rule Before Comparing Payoffs
A decision rule determines what “best” means when outcomes are uncertain. We recommend choosing the rule before calculating results, so your preferred option cannot determine the method. First, exclude options that breach a hard constraint, such as insufficient cash to meet existing obligations.
Maximin selects the option with the highest minimum payoff. It fits a decision where protecting the downside matters more than capturing possible upside. Its limitation is conservatism: a severe condition can dominate the choice even when you have little evidence about its likelihood.
Minimax regret selects the option with the smallest maximum missed payoff relative to other available choices. It fits a decision where you want to avoid being badly outperformed across plausible futures. Regret measures an opportunity gap, rather than the probability or absolute size of a loss.
When probabilities have credible support, expected payoff or expected utility may be more appropriate. Expected utility also accounts for how you value different outcomes. For consequential choices, establish the acceptable downside using decision frameworks for high-stakes decisions before allowing an attractive average return to justify the commitment.
A Worked Test: Full Launch, Partner Entry, or Wait
Suppose your SaaS team must choose how to enter a new market. For this example, assume the following net incremental cash outcomes over the next 12 months, after setup and operating costs. All figures are in thousands of dollars, and each option meets your initial feasibility requirements.
Payoff and regret tables are useful tools of decision-making when probabilities lack support. Keep the outcome measure and time horizon consistent across every option. The operating conditions below combine demand and market access, rather than assuming those uncertainties are independent.
Option
Strong demand, timely access
Weak demand, timely access
Market access delayed
Worst payoff
Full launch
240
-80
-160
-160
Partner entry
120
20
-40
-40
Wait
0
0
0
0
Read across each row to find its lowest payoff. Full launch has a minimum of negative $160,000, partner entry has negative $40,000, and waiting has zero. Maximin selects waiting, because zero is the highest of those minimum payoffs.
Waiting produces no incremental cash within this example’s horizon. Any actual holding costs must be deducted. The opportunity cost of missing a successful launch becomes visible through regret, which compares waiting with the best available choice under each condition.
Calculate regret by subtracting an option’s payoff from the best payoff in its column. Under strong demand with timely access, full launch produces $240,000 and partner entry produces $120,000. Partner entry therefore has $120,000 of regret under that condition.
Option
Regret: strong demand
Regret: weak demand
Regret: access delayed
Maximum regret
Full launch
0
100
160
160
Partner entry
120
0
40
120
Wait
240
20
0
240
Find the largest regret in each row, then select the smallest of those values. Minimax regret selects partner entry, with maximum regret of $120,000. That choice accepts some downside to reduce the largest missed opportunity across the tested conditions.
The rules disagree because they protect different interests. Waiting protects the worst cash outcome. Partner entry limits how far your choice could fall behind the best alternative, while full launch offers the highest upside.
These conclusions apply only to the options and conditions included. If full launch violates a cash limit, remove it before calculating regret. Recalculate each column against the best remaining feasible option, because an unavailable alternative should not determine your recommendation.
Gather Evidence That Could Change the Choice
Information has decision value when it can change your action or how you implement it. Identify the uncertain assumption behind the current recommendation, then specify what evidence would reverse it. This prevents open-ended research from becoming another form of overthinking.
For partner entry, verify whether the assumed loss is actually capped by the agreement. Obtain the payment schedule and cancellation terms, then check whether obligations continue after market access is delayed. A demand pilot can inform sales assumptions, but it cannot establish contractual rights or permission to operate.
Separate an evidence gap from a non-negotiable requirement. If operating permission is required, confirmation is a feasibility gate. If demand remains unclear, you may still choose a limited commitment, provided its downside fits your constraints and its results can inform the next decision.
AI can help expose missing assumptions, but generated estimates need verification. Apply the controls described in using artificial intelligence safely in decision-making when reviewing suggested consequences, especially where an unsupported claim could make an infeasible option appear attractive.
Before commissioning research, state which result would change the choice. If no plausible finding would alter your action, question the expense and delay. Without defensible probabilities, avoid presenting the expected value of information as a precise financial figure.
Keep a Decision Record You Can Update
A useful record for decision making under uncertainty states what would reverse your choice. Record the evidence behind each important payoff assumption, together with the owner responsible for checking it. Changing constraints then become a reason to update the comparison rather than restart the entire discussion.
Copy this template into your decision notes. Keep the distinction between observed facts and working assumptions visible, especially when several people contribute estimates. A review trigger should name an observable event, such as revised contract terms, rather than a vague instruction to reconsider later.
Decision and deadline:
Options, including waiting:
Outcome measure and time horizon:
Non-negotiable constraints:
Plausible adverse conditions to test:
Payoffs under each condition:
Evidence supporting the payoffs:
Assumptions still needing verification:
Decision rule and reason for choosing it:
Preferred option:
Evidence that would change the choice:
Next verification action and owner:
Review date or observable trigger:
For the market-entry example, a useful trigger is receiving a partner agreement that creates additional noncancelable obligations. That event changes the downside estimate directly. Update the affected payoffs before authorizing spending, even if the original comparison favored partner entry.
Frequently Asked Questions
Can I assign equal probabilities when I have no data?
You can use equal probabilities as an explicit modeling assumption, sometimes called the Laplace criterion. Missing evidence does not establish that outcomes are equally likely, so check whether the resulting recommendation depends on that assumption.
Can a good decision still produce a bad outcome?
Yes. A sound decision can lead to an unfavorable outcome because uncertainty remains after you choose. Evaluate the decision using the evidence available at the time, then investigate whether the outcome reveals an avoidable error or ordinary variability.
How do I compare options with nonfinancial consequences?
Use a common payoff measure only where the comparison is meaningful. Do not subtract a safety rating from a cash amount. Set acceptable boundaries for nonfinancial outcomes, then compare financial payoffs among options that satisfy those boundaries.
Can these rules handle decisions that unfold over time?
A single payoff table becomes limited when later actions depend on new observations. Sequential models, covered in Mykel Kochenderfer’s computational treatment of decision making under uncertainty, can represent those dependencies. They still require explicit assumptions about how conditions change and what you can observe.
Before choosing partner entry, verify the assumed downside against the agreement’s cancellation terms. To organize the verified alternatives, sign up for Lucid and use its AI-powered decision board to map their advantages, disadvantages, and future consequences.
