MAUT (Multi-Attribute Utility Theory)

MAUT is a ranking method for when you already hold a decision table and want the alternatives placed in a single order. Its output is a utility score for every alternative and the rank that score produces. Unlike TOPSIS or SAW, it does not work directly with raw or normalised values. It first builds a separate utility function for each criterion: a curve describing how the decision-maker values the figures on that criterion. It then passes every alternative through these functions and sums the results with weights. Keeney and Raiffa set the method out in 1976 as a systematic adaptation of utility theory to multi-attribute decision problems. It traces back to von Neumann-Morgenstern expected utility theory and counts as one of the founding frameworks in the decision-analysis literature.

MAUT's underlying idea is that "3 units" on one criterion does not carry the same value as "3 units" on another. Each criterion's own value must first be expressed, through the decision-maker's eyes, as a utility curve, and only then can these utilities be summed. Where classical SAW or TOPSIS normalises the raw value directly, MAUT inserts an interpretive layer in between: the utility function. This curve asks whether utility rises in step with the criterion's value, whether it saturates past a certain point, or whether it reflects caution in the face of risk. This is preference information drawn from the decision-maker; it does not fall automatically out of the data.

This idea carries a philosophical consequence: MAUT is compensatory and rests on an additivity assumption. When total utility can be written as the weighted sum of the individual criterion utilities, this is called the additive form. The form implicitly claims that criteria are preferentially independent: the value of a utility gain on one criterion does not depend on what happens on the others. Where this assumption fails, for instance where cost and quality are linked in the decision-maker's mind, the additive sum gives the wrong result. Keeney and Raiffa defined multiplicative and multilinear forms for such cases. This card describes only the additive form.

The method proceeds through four steps.

First, determining the single-attribute utility functions. A separate utility function u_j(x) is built for each criterion. The worst value the criterion can take receives a utility of 0, the best a utility of 1, and the values in between are curved according to the decision-maker's preference. This curve need not be linear. "Profit" might rise linearly for an investor, while "waiting time" for a hospital might deteriorate sharply past a certain threshold; such a curve is called convex or concave. Ideally these curves are obtained through structured interviews with the decision-maker, for instance using certainty-equivalent questions. This card uses a linear example for a formula-free explanation, but in real applications the curve's shape is an assumption that must be stated in the report.

Second, computing each alternative's utility on each criterion. The alternative's raw value on that criterion is passed through the criterion's utility function. The result is a utility score between 0 and 1.

Third, combining the multi-attribute utility in additive form. Each criterion's utility is multiplied by its own weight and the products are summed. For this summation to hold, the criteria must be preferentially independent in the decision-maker's judgement.

Fourth, ranking in descending order. Alternatives are arranged from the highest total utility score to the lowest.

The formulas behind each step are given on the DecisionMind method page; this card carries no formulas.

The utility score tells you how "valuable" an alternative is found on the decision-maker's own utility scale. A score of 0.65 does not mean "65 per cent good" or "65 per cent likely to be the best." It only shows where the alternative sits between 0 (worst on every criterion) and 1 (best on every criterion). If the utility functions were derived independently of the decision-maker's alternatives, anchored only to that criterion's best and worst possible extremes, this score can be compared across different analyses, because the scale is fixed. But if the utility function is instead built from the largest and smallest values among the alternatives on hand, as in this card's example, the position changes: the score becomes dependent on the alternative set, just as with TOPSIS, and cannot be compared. The report must state clearly which route was taken.

Thus instead of writing:

"MAUT showed that this alternative's true value is 0.65"

the report should read:

"With these utility functions and these weights, the highest total utility belongs to this alternative; whether the utility scale is fixed or built relative to this alternative set is as follows"

Classical MAUT works with crisp data: one number per cell. If your data is fuzzy, given as a range, or contradictory across experts, DecisionMind holds four MAUT family members alongside the base method, including fuzzy, intuitionistic fuzzy and plithogenic extensions. Which extension fits which data situation is explained on the relevant data-type cards.

You need alternatives in rows, criteria in columns, one number per cell, no empty cells; direction information for every criterion; a utility function for every criterion (at minimum its endpoints: the utility corresponding to the worst and best values); and criterion weights that sum to 1. MAUT does not produce weights, it asks for them; you can derive weights from subjective methods such as AHP, BWM or SWARA, or from objective methods such as Entropy or CRITIC. A minimum of two alternatives and two criteria is required, and three to twelve criteria work comfortably. Because eliciting utility functions takes time, the elicitation burden grows as the number of criteria increases.

MAUT is a sound choice where the decision-maker's preference on a criterion may not be linear, for instance where utility falls sharply or saturates past a threshold, and where this curve can be elicited. It brings into the model, explicitly, a non-linear preference that methods normalising raw values alone would overlook. Its typical territory includes alternative evaluation, supplier selection, health policy and investment decisions; it is meaningful anywhere risk attitude matters.

There are two cases where it should not be used. First, if the criteria are dependent in the decision-maker's preference, meaning the value of a utility gain on one criterion changes according to what happens on another, the additive sum is invalid and a multiplicative or multilinear form is required. Second, MAUT is unsuitable where there is no time or expert access to elicit utility functions; a simple weighted sum (SAW) or a method that normalises directly (TOPSIS) demands less data in that case.

Criterion utility is non-linear, the curve can be elicited, criteria are independent → MAUT

No time to elicit a utility curve, a quick weighted sum suffices → SAW

Criteria are dependent in the decision-maker's preference → multiplicative/multilinear utility forms

No compromise allowed on one criterion, sub-threshold alternatives must be screened out → screening first, then ranking

Not a ranking but weights are needed → AHP, BWM, SWARA (subjective) · Entropy, CRITIC (objective)

MAUT's fundamental advantage is that it can bring the decision-maker's risk attitude and non-linear preferences directly into the model. It captures situations where "rising from 3 to 4" on a criterion is not counted as the same utility gain as "rising from 8 to 9." Its theoretical foundation is sound: it rests on the von Neumann-Morgenstern expected-utility axioms and offers a framework consistent with preferences under uncertainty (Keeney and Raiffa, 1976). Because each alternative's utility on each criterion can be seen separately, the result is explainable; how much each criterion contributes to total utility can be traced step by step.

Its limitations largely follow from its assumptions. First, the additive form requires preferential independence; where the criteria are linked in the decision-maker's judgement this sum gives the wrong result and a multiplicative form is needed instead (Keeney, 1974). Second, deriving utility functions correctly is costly, requiring structured interviews, certainty-equivalent questions or lottery comparisons. Proceeding without this effort, on a "roughly linear" assumption, makes MAUT practically indistinguishable from methods that simply normalise raw values (Dyer, 2005). Third, there is the assumption of full compensation: a serious weakness on one criterion can be papered over on the utility scale by others. Fourth, practitioners have observed that MAUT has not become as widespread in the field as AHP or TOPSIS, owing to its elicitation burden; practical adoption has stayed limited despite its theoretical maturity (Wallenius et al., 2008).

The most common mistake is presenting plain min-max normalisation as a "utility function" without ever eliciting one. This skips MAUT's theoretical foundation, namely preference information drawn from the decision-maker, and leaves the result dependent on the alternative set. A second mistake is applying the additive form unquestioningly when a known dependency exists between criteria, for instance where cost and quality move together; preferential independence must be checked. A third mistake is reading the utility score as a probability or a percentage. A fourth is assigning equal weights without justification and treating this as "neutrality"; equal weight is itself a preference that needs defending. A fifth is assuming linearity for a criterion known to have a non-linear utility curve, for instance a risk indicator that deteriorates sharply past a threshold.

The governing principle is this:

A MAUT result is a summary not only of the input data but of the decision-maker's utility curves and the preferential-independence assumption. If these curves and this assumption are contested, the result is contested too, and the report must show this.