Decision Analysis Calculator

What this tool decides

Decision analysis is the discipline of choosing well when the result of your choice depends on events you cannot control.

You lay the problem out as a payoff table: the rows are the alternatives you can choose between, the columns are the states of nature that might occur, and each cell is the payoff, a profit or a cost, that follows from that combination.

The trouble is that no single alternative is best in every state; the aggressive choice wins big if the good state occurs and loses badly if it does not, while the safe choice gives up upside for protection. Decision criteria are the formal rules that turn that tension into a recommendation, each encoding a different attitude toward the unknown.

This tool takes one payoff table and runs every standard criterion on it at once.

Under uncertainty, when you cannot put numbers on how likely each state is, it reports the optimist’s maximax, the pessimist’s maximin, the compromise Hurwicz criterion with an optimism dial you control, the equal-likelihood Laplace criterion, and the minimax-regret rule, and it shows the full regret matrix they rest on.

Under risk, when you can assign a probability to each state, it computes the expected monetary value of every alternative, the expected opportunity loss, and the expected value of perfect information, the ceiling on what a perfect forecast would be worth. It handles profit tables and cost tables alike, and everything runs in your browser with nothing stored.

The value of seeing all the criteria together is that it reveals how much your choice depends on your stance rather than on the numbers alone. When every criterion points to the same alternative, the decision is robust and easy. When they disagree, as they often do, the disagreement is itself the insight: it tells you the right choice hinges on how optimistic you are or how much downside you can bear, and it focuses the conversation on that question rather than on the arithmetic, which the tool has already done.

How to use this calculator, step by step

Begin by choosing whether the payoffs are profits, where larger is better, or costs, where smaller is better, and whether you are deciding under uncertainty or under risk. Then type the payoff table: one alternative per row, and within each row the payoff for each state of nature, separated by spaces or commas, with the states in the same order across rows. The tool opens with a solved three-by-three profit example, a plant-sizing decision across low, medium, and high demand, so you can see a complete set of recommendations before you change anything.

You can name the alternatives and the states so the output reads in your own terms rather than as generic labels. The Hurwicz coefficient of optimism, alpha, is a dial from zero to one: nudge it toward one to lean optimistic and toward zero to lean cautious, and watch the Hurwicz recommendation move. If you switch the decision mode to risk, a box appears for the state probabilities, which must add up to one; enter them and the tool adds the expected-value analysis. Everything recomputes live as you edit, so exploring how the answer responds to your inputs is immediate.

The result panel leads with a headline recommendation, then a table listing what each uncertainty criterion chooses and the value behind it, with the chosen alternative highlighted. In risk mode a second table gives the expected monetary value and expected opportunity loss of every alternative and states the expected value of perfect information. Below those, the full regret matrix shows the opportunity loss in every cell, and a chart compares the alternatives on the deciding measure. You can download the analysis as a PDF or CSV or share it, all locally.

Reading the payoff table and the criteria

The payoff table is worth reading carefully because every recommendation flows from it. Scanning a row tells you how one alternative performs across all the futures; a row that is high everywhere is a dominant choice you can pick without further thought, while a row that swings from very high to very low is a gamble whose appeal depends entirely on your read of the states. Scanning a column tells you which alternative is best if that particular state occurs, which is the raw material for the regret calculation. The spread within the table, how much the payoffs move as the state changes, is what makes the decision hard and what the criteria are there to resolve.

Each uncertainty criterion answers a different question about that table. Maximax asks which alternative has the best best-case and picks it, ignoring downside, the pure optimist. Maximin asks which alternative has the least-bad worst-case, protecting the floor, the pure pessimist.

Hurwicz blends those two extremes in the proportion you set with alpha, letting you sit anywhere between them. Laplace treats the states as equally likely and picks the best average, using the whole table rather than just its extremes.

Minimax regret asks a subtler question, which alternative minimises the worst regret you would feel once the state is revealed, and it is the one criterion that looks at how each choice compares to what would have been best rather than at the payoff alone.

Because the criteria encode different attitudes, they need not agree, and this calculator makes the pattern of agreement visible at a glance. If the cautious and the optimistic criteria both point to the same alternative, that choice is safe across attitudes and you can act with confidence. If they split, the highlighted cells show exactly which alternatives are in contention and under which stance, turning an abstract disagreement into a concrete short list. Reading the criteria as a set, rather than trusting any single one, is the way experienced analysts use a payoff table.

Decision under uncertainty versus decision under risk

The single most important distinction in this field is whether you can assign probabilities to the states of nature. When you cannot, you are deciding under uncertainty, and the relevant tools are the maximax, maximin, Hurwicz, Laplace, and minimax-regret criteria, each of which makes a recommendation without any probabilities at all. They differ only in the attitude they encode, which is why looking at several of them together is more honest than committing to one, and why this calculator reports all of them side by side.

When you can assign probabilities, you are deciding under risk, and the natural tool is expected monetary value, which weighs each payoff by how likely its state is and averages. EMV gives a single recommendation and is the standard in business precisely because it uses all the information you have.

Alongside it, the expected opportunity loss measures the average regret of each alternative, and its smallest value equals the expected value of perfect information, the ceiling on what you should pay to remove the uncertainty.

Switching this calculator from uncertainty to risk keeps your payoff table and simply adds the probability box and the expected-value results, so moving between the two views costs nothing.

In practice the two modes work together. You might start under uncertainty because you have no probabilities, use the criteria to understand how the choice depends on attitude, then gather enough evidence to estimate probabilities and switch to risk for a sharper recommendation and a value on further information. The uncertainty criteria never become useless once you have probabilities; they remain a check on how sensitive the decision is to your assumptions, and a decision that survives both views is one you can defend.

Five worked examples you can follow

Example 1: the criteria disagree

The calculator opens with a plant-sizing profit table where the criteria genuinely split. The optimist’s maximax points to the large plant on the strength of its high-demand payoff, the pessimist’s maximin points to the small plant because its worst case is safest, and Laplace and minimax regret both settle on the medium plant as the balanced choice. This disagreement is the point: it tells you the right size depends on your read of demand and your appetite for risk, and it frames the real conversation rather than hiding it behind a single number.

Example 2: turning the optimism dial

Keep the same table and move the Hurwicz alpha from zero toward one. At alpha zero the criterion coincides with the cautious maximin choice; as you raise it the recommendation shifts toward the aggressive alternative, and somewhere in between it crosses over. Finding that crossover tells you how optimistic you would have to be to justify the bolder choice, which is often a more useful thing to know than any single recommendation, because you can judge whether your genuine outlook is above or below that threshold.

Example 3: adding probabilities

Switch the mode to risk and enter probabilities for low, medium, and high demand. The calculator now reports the expected monetary value of each alternative and picks the highest, and in this example the medium plant wins on EMV. The expected-value view collapses the earlier disagreement into one recommendation because it uses the probabilities the uncertainty criteria lacked, showing how information sharpens a decision that attitude alone left open.

Example 4: the value of a forecast

With probabilities entered, read the expected value of perfect information the calculator reports. It is the gap between what you could earn if you always knew demand in advance and what you earn choosing the best single alternative now. That number is the most you should pay for a perfect market study, and any real study, which is never perfect, is worth less. Seeing it in currency terms turns the vague question of whether to invest in research into a concrete budget ceiling.

Example 5: a cost table

Switch the payoff type to cost and enter a table of costs, for example the cost of three supply strategies under different demand scenarios. Every criterion flips: the optimist now seeks the lowest possible cost, the pessimist protects against the highest, and EMV picks the lowest expected cost. The regret matrix is computed against the cheapest option in each state. This shows that the same framework handles minimizing losses just as cleanly as maximizing gains, with no change in how you enter the data.

Three expert tips for a sound decision

The mathematics behind the results

Each criterion is a short calculation on the payoff table. For a profit table, maximax takes the maximum of each row and then the alternative with the largest of those row-maxima, while maximin takes the minimum of each row and then the alternative with the largest of those row-minima. The Hurwicz value of an alternative is alpha times its best payoff plus one minus alpha times its worst, and you choose the largest; at alpha one it reduces to maximax and at alpha zero to maximin. Laplace averages each row and chooses the largest average. For a cost table each of these flips to its minimizing mirror, which the calculator handles automatically from the payoff-type switch.

The regret matrix underlies both minimax regret and the risk analysis. For a profit table, the regret in a cell is the best payoff available in that state of nature minus the payoff you actually get there, so it measures the opportunity you forgo by not having chosen the state’s best alternative; every regret is zero or positive. Minimax regret takes the largest regret in each row and chooses the alternative whose largest regret is smallest. For a cost table the regret is your cost minus the lowest cost available in that state, and the logic is the same.

Under risk the probabilities drive three numbers. The expected monetary value of an alternative is the sum over states of the state probability times the payoff, and you choose the best EMV. The expected opportunity loss is the same weighted sum applied to the regret matrix, and the alternative with the smallest expected opportunity loss is always the same as the best-EMV alternative.

The expected value of perfect information is the expected payoff under perfect foresight, the probability-weighted sum of the best payoff in each state, minus the best EMV, and it equals that smallest expected opportunity loss, a tidy identity the calculator displays so you can verify it.

All of these are elementary once the table is in place, which is why decision analysis is powerful out of proportion to its arithmetic.

Where decision analysis is used

Payoff-table analysis appears wherever a choice must be made before an uncertain event resolves. In capacity and capital planning it sizes plants, warehouses, and fleets against uncertain demand.

In product and marketing decisions it weighs launching, delaying, or cancelling against uncertain reception, and it prices the market research that would reduce that uncertainty through the expected value of perfect information.

In procurement and supply it compares sourcing strategies against uncertain prices or disruptions, and in finance it frames investment and insurance choices against uncertain returns and losses. Any decision that can be written as alternatives against states of nature fits the frame.

The method is also a teaching staple because it isolates the role of attitude toward risk so cleanly. By showing that the optimistic, cautious, and expected-value criteria can each point to a different alternative from the same table, it makes vivid that there is often no single correct choice independent of the decision maker’s stance, only choices that are correct given a stance. That lesson carries into more elaborate methods, decision trees that chain several decisions and events, and utility theory that replaces money with a value function capturing risk preference, both of which build directly on the payoff table this calculator analyses.

Within the operations research toolkit, decision analysis pairs naturally with the optimisation and stochastic models.

Where linear programming finds the best decision when the world is known and constraints bind, decision analysis handles the case where the world is uncertain and the choice is among a few discrete alternatives.

And where a Markov chain models how a system evolves through states over time, a payoff table captures a single decision against a one-shot uncertain state. Return to the Operations Research hub for the full set of models.

Building a real payoff table

Turning a real decision into a payoff table is mostly the discipline of naming the alternatives and the states well. The alternatives should be the genuine, distinct options you are choosing among, few enough to compare and each one a course of action you could actually take. The states of nature should be the uncertain outcomes that most affect how the alternatives perform, defined so that they are collectively exhaustive, covering every future that matters, and mutually exclusive, so exactly one of them will occur. Two or three well-chosen states usually capture the essential uncertainty; too many make the payoffs hard to estimate and the table hard to read.

Each payoff should be the net result, profit or cost, of that alternative under that state, expressed in one consistent unit. Estimating these is the real work, and it is worth being honest that they are estimates; a payoff table with false precision invites false confidence. Where a payoff depends on several factors, compute it separately and enter the total, and keep the sign convention consistent, positive for gains and negative for losses in a profit table, so the criteria interpret them correctly. If two objectives matter at once, say profit and risk, you must fold them into a single number per cell before the table can be analysed, because every criterion optimises one measure.

Finally, decide up front whether you can assign probabilities. If solid evidence supports them, enter them and use the expected-value analysis, but resist inventing probabilities to make the problem look tidier than it is; false probabilities are worse than honest uncertainty. If you truly cannot assign them, stay in uncertainty mode and let the spread of criteria show how the choice depends on attitude. Being clear about which situation you are in keeps the analysis honest and points you to the results that actually apply.

When the payoff table is the wrong tool

The payoff table is deliberately simple, and its simplicity marks its limits. It models a single decision made once against a single uncertain event; when a decision unfolds as a sequence, choose, then observe, then choose again, a decision tree is the right tool, because it can represent the later choices that the earlier ones open up and fold their values back. Trying to flatten a multi-stage problem into one table loses the structure that makes it interesting, so recognising when your problem is really a sequence is the first check.

A second limit is the assumption that money, or whatever the payoff measures, captures value linearly. For large stakes relative to the decision maker’s resources, a guaranteed amount is often worth more than a gamble with the same expected value, a preference that expected monetary value cannot express.

Utility theory addresses this by replacing payoffs with utilities that bend to reflect risk aversion, and when the stakes are large enough to threaten the enterprise, an EMV recommendation should be sanity-checked against that risk attitude rather than followed blindly. A third limit is dependence between the decision and the state: the framework assumes the states’ likelihoods do not depend on which alternative you choose, and where your choice would change the odds, a richer model is needed.

None of these caveats undermines the payoff table on the many decisions that are genuinely single-shot, money-valued, and independent; they simply mark where a decision tree or utility analysis earns its extra complexity.

Reading beyond the recommendation

A payoff-table analysis gives more than a winner, and the extra information is often where the real decision is made. The most important habit is to look at the pattern across criteria rather than any single recommendation. When the criteria converge, the decision is robust and the analysis mainly confirms it; when they diverge, the divergence tells you precisely what the decision turns on, usually your attitude to risk or your read of the states, and that is the question worth debating. The tool lays the criteria side by side so this pattern is immediate rather than something you have to assemble.

The regret matrix rewards a second look. A cell with a large regret marks a combination you would most regret walking into, and an alternative with a small maximum regret is one that never leaves you far from the best you could have done, which is a form of robustness distinct from a high average. Scanning the regret matrix often surfaces an alternative that no single payoff criterion crowned but that is quietly safe across every state, and such an option is frequently the wise real-world choice even when a bolder one wins on paper.

Finally, treat the payoffs and any probabilities as estimates and test how the recommendation responds to changing them. If nudging one payoff or shifting a probability flips the recommendation, the decision is sensitive there and that input deserves more care before you commit; if the recommendation holds across reasonable variations, you can act with more confidence. This calculator recomputes instantly, so running a handful of such what-ifs takes seconds and tells you how much the conclusion really rests on the exact numbers, which is usually more valuable than the first headline it produced.

A brief history of the idea

The formal criteria in this calculator crystallised in the middle of the twentieth century, as statisticians and economists worked to put decision making under uncertainty on a rigorous footing. Abraham Wald introduced the pessimistic maximin idea in his work on statistical decision functions in the 1940s, framing choice as a game against a hostile nature.

Leonard Savage proposed the minimax-regret criterion in 1951, shifting the focus from payoff to opportunity loss, and around the same period Leonid Hurwicz offered the optimism-weighted compromise that bears his name.

The equal-likelihood idea reaches back much further, to the principle of insufficient reason associated with Laplace in the eighteenth and nineteenth centuries.

The probabilistic side grew from the same era’s advances in expected-value reasoning and Bayesian statistics, and it was gathered into a practical discipline for managers by Howard Raiffa and Robert Schlaifer at Harvard in the early 1960s, who coined much of the language of decision analysis, including the expected value of perfect information.

What began as competing philosophical answers to a hard question, how to choose when you do not know what will happen, became a standard toolkit taught in every operations management and business statistics course.

That a handful of one-line rules on a small table can frame decisions from plant sizing to product launches, and can even price the information that would resolve the uncertainty, is why the payoff table remains a durable and widely taught model.

From table to shared decision

One underrated benefit of laying a choice out this way is that it makes a decision discussable. A group arguing in the abstract about whether to be bold or cautious rarely converges, but the same group looking at a shared table can locate exactly where they disagree: on the payoffs, on the states, on the probabilities, or on the attitude to risk. Each of those is a different conversation, and separating them is half the battle. The numbers stop being a rhetorical weapon and become a common object everyone can inspect and challenge, which tends to raise the quality of the debate and leave a clear record of why a choice was made.

That record matters after the fact as much as before it. When the uncertain state finally resolves and the outcome is known, a documented payoff table lets a team review the decision on its merits rather than its result, distinguishing a good decision that met bad luck from a genuinely poor choice. Judging decisions by their process rather than only their outcome is a habit that compounds over time, and a simple archived table is one of the cheapest ways to build it into how an organisation works.

Input format and quick reference

Choose profit or cost and uncertainty or risk, then type the payoff table one alternative per row with one payoff per state, separated by spaces or commas. Optionally name the alternatives and states, set the Hurwicz alpha, and in risk mode enter probabilities that add up to one. The reference below explains each result.

Frequently asked questions

Sources, disclaimer and editorial transparency

This calculator applies standard operations-research decision criteria to a payoff table: maximax, maximin, minimax regret (Savage), Hurwicz with an adjustable coefficient of optimism, and Laplace under uncertainty, and expected monetary value, expected opportunity loss, and the expected value of perfect information under risk, with full support for profit and cost objectives. This calculator and guide are created and reviewed by the OpsCalculators team; see our Editorial Policy for how each tool is researched, built, and tested.

Results are accurate estimates for planning and education, not certified financial or engineering advice, and they assume your states of nature are collectively exhaustive and mutually exclusive, that payoffs are on one consistent scale, and, in risk mode, that the probabilities you enter are sound. For large stakes, sanity-check an expected-value recommendation against your risk attitude. See our full Disclaimer. OpsCalculators.com is operated by MAFHH INTERNATIONAL LTD. Your data is processed in your browser and never stored; see our Privacy Policy.