What is the Analytic Hierarchy Process?

An introduction to the AHP

The Analytic Hierarchy Process (AHP) is a structured method for making decisions that involve multiple competing criteria. It works by breaking a decision into a hierarchy (goal, criteria, alternatives), asking the decision maker to compare elements two at a time on a 1–9 scale, and mathematically deriving priority weights from those comparisons. The result is a defensible, transparent ranking of the alternatives, together with a built-in check on the logical consistency of the judgments behind it.

The AHP was developed in the 1970s by Thomas L. Saaty, a mathematician and operations researcher at the Wharton School of the University of Pennsylvania. His key insight was to combine mathematical structure with how humans naturally think about preferences. Rather than asking people to assign abstract percentage weights to a list of criteria, the AHP asks a series of simpler questions: given two factors, which matters more, and by how much? Saaty tells the origin story himself in a series of short videos.

Since its introduction, the AHP has become one of the most widely used methods for multi-criteria decision analysis. It is applied across business, government, healthcare, and engineering, is taught in business schools worldwide, and features in quality management methodologies including Six Sigma. You can see it applied to real published decisions in our AHP case studies, or open a live worked AHP example: family car selection in decisionpoint.io without signing in.

How does the AHP structure a decision?

The AHP structures a decision as a hierarchy: a single goal at the top, the criteria that matter to that goal below it, optional subcriteria below those, and the alternatives under consideration at the bottom. Decomposing the problem this way means no single judgment ever has to weigh everything at once; each comparison is made within one small, homogeneous group of elements. The figure at the top of this page shows this structure for a smartphone choice, the example worked through in the sections below.

• Goal: the decision to be made, stated as a single objective (e.g. "choose the best supplier")
• Criteria: the factors that matter for this decision (e.g. cost, quality, delivery reliability)
• Subcriteria (optional): finer-grained factors grouped under a criterion (e.g. cost might split into purchase price and running cost); each group is compared within its parent
• Alternatives: the options being ranked

Within each level, elements are compared pairwise: first the criteria against each other with respect to the goal, then the alternatives against each other with respect to each criterion. A group of n elements requires n(n−1)/2 comparisons. Best practice keeps each group to around seven elements or fewer, both to limit the number of comparisons and because human judgment degrades when more factors are compared at once. For a complete real hierarchy with 10 criteria and 6 alternatives, see the Wikipedia car-buying case study.

How does the 1–9 comparison scale work?

Pairwise comparisons in the AHP are expressed on a 1–9 ratio scale known as the Saaty scale, or fundamental scale. Each judgment answers one question: "With respect to [this criterion], how much more important is element A than element B?" A value of 1 means the two elements contribute equally; 9 means the evidence favouring one over the other is of the highest possible order.

The scale is reciprocal: if A is judged 3 times as important as B, then B is automatically 1/3 as important as A. Only one judgment per pair is needed. Fractional values between the whole points (e.g. 3.5, or the reciprocal 1/3.5) are valid on the scale, and decisionpoint.io supports them for judgments that fall between the standard definitions.

Why stop at 9? Saaty's main reason is cognitive: people can only make reliable qualitative distinctions across roughly nine gradations, and elements being compared should be homogeneous, within about one order of magnitude of each other. When elements genuinely differ by more than an order of magnitude, they belong at different levels of the hierarchy rather than at the extremes of the scale. There are mathematical reasons too: the calculation that turns judgments into weights is most reliable when the judgments stay within a modest range, as explained in the eigenvector section of the mathematics page. Pairwise comparison itself reflects the same cognitive principle: people are more accurate comparing two things directly than assigning weights to ten things at once.

How are AHP weights calculated?

The AHP derives priority weights from pairwise judgments by considering all of them at once. The judgments for a group of elements are arranged in a comparison matrix, a table with one row and one column per element, and the method finds the single set of weights that best fits every judgment in the table. If the judgments were perfectly consistent, the weights would reproduce them exactly. When they are mildly inconsistent, as human judgments always are, the method returns the best compromise, so no single judgment dominates the result.

The calculation behind this is known as the eigenvector method, and is typically calculated using software, such as decisionpoint.io. For hand calculation, taking the geometric mean of each row of the matrix and scaling the results to sum to 1 gives a close approximation. For a detailed dive into how to perform these calculations, see the companion Mathematics of the AHP article.

Worked example: weighting three criteria

Suppose you are choosing a smartphone using three criteria: Camera, Price, and Battery. You judge Camera moderately more important than Price (3), Camera strongly more important than Battery (5), and Price moderately more important than Battery (3). The comparison matrix and resulting weights are:

Camera carries 63.7% of the weight, Price 25.8%, and Battery 10.5%.

Worked example: scoring the alternatives

The same procedure is applied to the alternatives under each criterion: the three phones are compared pairwise with respect to Camera, then Price, then Battery, and each set of comparisons yields a set of local priorities by the same method. The judgments and resulting local priorities are (a fractional value means the second phone is preferred, so 1/5 under "X vs Y" means Phone Y is strongly preferred to Phone X):

All three sets of judgments are comfortably consistent; the consistency ratio is explained below. Final scores are then synthesised by weighted sum: each alternative's local priorities are multiplied by the corresponding criterion weights and added up:

Phone Y ranks first, driven by its camera performance under the dominant criterion. To see these calculations performed live on a full decision, open the worked AHP example: family car selection in decisionpoint.io, which shows every comparison matrix, weight, and final score.

What is a consistency ratio?

The consistency ratio (CR) measures how logically coherent a set of pairwise judgments is, and by convention a CR of 0.10 or less is considered acceptable. Perfectly consistent judgments hang together: if A is 3 times B and B is 3 times C, a perfectly consistent judge would rate A as 9 times C. Human judgments rarely achieve this, and the AHP does not require them to; the CR quantifies how far they deviate.

The CR is calibrated against randomness: it compares the inconsistency in your judgments with the inconsistency typically found in judgments made entirely at random for a group of the same size. A CR of 0.10 therefore means the judgments are 10% as inconsistent as random noise.

In the smartphone example above, the judgments 3 (Camera vs Price) and 3 (Price vs Battery) would imply 9 for Camera vs Battery under perfect consistency, but 5 was entered, so the judgments are mildly inconsistent. Their CR is about 0.03, comfortably below the 0.10 threshold. Had the CR exceeded 0.10, the remedy is to revisit the most inconsistent judgments, not to force artificial precision: Saaty regarded some inconsistency as natural and even informative. (He also suggested tighter thresholds of 0.05 for groups of three elements and 0.08 for groups of four, where random inconsistency is harder to come by.) decisionpoint.io computes the CR continuously and can point you at the specific judgments that contribute most to it. How the CR is calculated, with the random index table it depends on, is on the mathematics page.

How does the AHP work for groups?

Group AHP combines the pairwise judgments of multiple participants into a single group result, most commonly by taking the element-wise geometric mean of their judgments. This replaces debate-driven consensus, where seniority and personality can skew outcomes, with a mathematical aggregation in which every participant's view counts equally and disagreement remains visible.

The geometric mean is used rather than the familiar arithmetic mean because it respects the reciprocal nature of the scale. If one participant judges A vs B at 3 and another at 1/3, the geometric mean is exactly 1, correctly reflecting a balanced group; the arithmetic mean would not be. The reasoning, and the alternative of averaging each participant's final weights instead of their judgments, are covered on the mathematics page.

decisionpoint.io implements group AHP by aggregating judgments with the geometric mean, in two modes: workshop mode for panels iterating towards consensus, and survey mode for collecting independent anonymous judgments at scale. See the group AHP feature announcement for details, or open the live examples: a group AHP workshop example: selecting a customer support platform and a group AHP survey example: community wind farm site selection. For contrast, the New York Times ID format case study shows a team applying the AHP by debating each judgment to consensus instead of aggregating mathematically.

Can the AHP be used without alternatives?

Yes. A criteria-only AHP analysis stops after weighting the criteria, and the weights themselves are the deliverable. This is appropriate whenever the question is "how much does each factor matter?" rather than "which option wins?": allocating a budget across programmes, weighting the components of an evaluation rubric or scoring model, prioritising requirements or risk factors, or setting the weights of key performance indicators.

Criteria-only analysis is also the right structure when the alternatives cannot be enumerated up front, for example when proposals or candidates arrive continuously over time. In that case the AHP-derived weights are paired with ratings mode (described below): each incoming alternative is scored against fixed intensity scales, so new options can be evaluated without re-running any pairwise comparisons. The mathematics is unchanged; the hierarchy simply has no alternatives level, and the consistency check applies to the criteria comparisons exactly as described above.

What is the AHP used for?

The AHP is used wherever a consequential choice must balance several competing factors and be justified to others. Three application areas recur constantly in practice:

• Supplier and vendor selection: procurement teams weigh cost, quality, delivery, and risk to rank suppliers, as in this live example of a panel selecting a customer support platform.
• Site and facility selection: siting decisions balance technical suitability against environmental and community factors, as in the community wind farm site selection survey.
• Healthcare priority setting: hospitals and health systems use the AHP to weight clinical, economic, and patient-centred criteria when prioritising treatments, technologies, and research investments.

The same structure handles engineering and technology choices (the New York Times' selection of an ID format), major procurement (the Brazilian Navy's warship selection), and personal decisions (the classic car-buying example). Browse all of these in the case studies section.

What are the strengths and limitations of the AHP?

The AHP's main strengths are transparency and cognitive fit: it reduces a complex decision to a series of simple two-way judgments, produces a fully traceable ranking, and flags incoherent judgments via the consistency ratio. Its main limitations are the growth in the number of comparisons as the hierarchy widens and its sensitivity to how the problem is framed.

• It breaks complex problems into manageable pieces
• It combines subjective judgment with structured analysis
• Results are transparent, so you can trace exactly how the final ranking was derived
• It extends naturally to group decisions, synthesising input from multiple stakeholders (see group AHP)
• The consistency check keeps judgments logically grounded

The limitations worth planning around:

• The number of pairwise comparisons grows quadratically: ten criteria means 45 comparisons for the weights alone, which can become tedious without software support.
• The quality of results depends heavily on thoughtful criteria selection; poorly chosen or overlapping criteria produce misleading rankings.
• Best practice keeps each comparison group to around seven elements or fewer, reflecting cognitive limits on how many factors humans can reliably compare.

What are the main variants of the AHP?

Several variants extend the core method to situations its standard assumptions do not cover:

Analytic Network Process (ANP) is Saaty's own generalisation of the AHP that handles interdependencies and feedback between criteria. Where the AHP assumes criteria are independent, ANP allows for more complex relationships: useful when, for example, price affects perceived quality, or when the alternatives influence which criteria matter.

Fuzzy AHP combines the AHP with fuzzy set theory to handle uncertainty in judgments. Instead of precise numerical comparisons, decision makers can express preferences as ranges (e.g. "between moderately and strongly important"), accommodating the vagueness inherent in human judgment.

AHP-express is a simplified approach that reduces the number of required comparisons by comparing every element against a single reference element, making the method more practical for time-constrained applications while preserving the core logic.

Ratings mode (also called absolute measurement) replaces pairwise comparison of alternatives with scoring each alternative against predefined intensity scales (e.g. Excellent/Good/Average/Poor), whose intensities are themselves weighted by pairwise comparison. It is the natural companion to criteria-only analysis when there are many alternatives or when alternatives arrive over time.

How do you get started with the AHP?

The most important first step is not choosing software but clearly defining the goal and identifying the criteria that genuinely matter; time spent on this foundation pays dividends throughout the process. The AHP can be done with pen and paper for small decisions, but software makes it practical at real-world scale. Tools range from desktop applications like SuperDecisions, phone apps such as Decision Mentor, and purpose-built web applications such as this one, decisionpoint.io, designed for ease of use and collaboration.

For a fully worked example that goes beyond the smartphone walkthrough above, see the Wikipedia car-buying case study: a complete AHP decision with 10 criteria, 6 alternatives, 133 pairwise comparisons, and a full sensitivity analysis. You can also open live, interactive examples in decisionpoint.io without signing in: a worked AHP example: family car selection and a group AHP survey example: community wind farm site selection. Both show the full hierarchy, every pairwise comparison, and the calculated results.

And when you are ready for the primary sources — Saaty's original papers, the major critiques, and his recorded lectures — the AHP Library collects them with plain-language notes and verified links.

References