How to use it 01 List the criteria and edit only the upper triangle of the comparison matrix. 02 Use 1–9 for "how many times more important", and fractions such as 1/3 for the reverse direction. 03 Read the weight bars and the consistency block: λmax, CI, CR and the random index RI. 04 Turn on alternatives to add a second layer and read the global ranking.
Worked readouts Four criteria — reliability, cost, support, speed — compared at 2, 4, 2 / 3, 1 / 1/2:
Weights 43.8% / 24.4% / 9.9% / 21.9% ; λmax 4.0206 , CI 0.0069 , CR 0.0076 against RI 0.90 — comfortably consistent.
The three-item demo (price/location/area at 1/3, 5 and 7) reads 27.9% / 64.9% / 7.2% with CR 0.056 .
A cyclic matrix — A beats B 9, B beats C 9, C beats A 9 — collapses to 33.3% each with CR 6.1303 : the warning fires and the weights are not usable as stated.
Where weights come from The comparison matrix is square and reciprocal: aⱼᵢ = 1/aᵢⱼ with diagonal 1. The weight vector is the principal eigenvector; the row geometric mean is available as an alternative estimate. λmax ≥ n with equality for a perfectly consistent matrix, CI = (λmax − n)/(n − 1), and CR = CI/RI uses Saaty's random index for the matrix size. A CR at or below 0.1 is the working threshold.
Limits
Judgements, not measurements. Weights inherit the comparisons; a passing CR only says the comparisons do not contradict each other, not that they are right.
Size and reciprocity. Up to ten items per comparison; a matrix that breaks reciprocity beyond a 2% rounding tolerance is rejected. What it does not do. It does not score alternatives from data or estimate costs; scoring belongs to the weighted matrix . This page provides general information and scenario estimates based only on the inputs you provide. It is not investment, tax, legal, or other professional advice and does not recommend any rate, product, or institution.
Equal weights are not "no preference" The cyclic matrix returns 33.3% for every criterion — not because the judge was indifferent, but because the cycle is symmetric and the eigenvector is flat. The CR of 6.1303 is the signal that the numbers, not the weights, need fixing. The two estimation methods also drift apart as inconsistency grows: on the four-criterion example the eigenvector and geometric-mean weights differ slightly (43.79% against 43.87% for the first criterion), while on a consistent matrix they agree exactly.
Where it is used Deriving weights before scoring Compare criteria pairwise, check CR, then carry the weights into the weighted matrix and score the alternatives.
Two-level ranking Turn on alternatives, compare them under each criterion, and read the global scores as a weighted average of the local weights.
Privacy Comparisons, weights and consistency statistics are computed in the browser; nothing is uploaded.
References
Saaty, T.L., A scaling method for priorities in hierarchical structures , Journal of Mathematical Psychology 15(3), 234–281, 1977, doi.org (访问日期:2026-10-07)— the 1–9 scale, priority vectors and consistency.
Wikipedia, Analytic hierarchy process , en.wikipedia.org (访问日期:2026-10-07)— random index and the CR ≤ 0.1 rule.
Wikipedia, Multiple-criteria decision analysis , en.wikipedia.org (访问日期:2026-10-07)— the wider family of weighting methods.
Calculators in this hub
Hand-picked tools, one click away. The mini versions compute live and carry your values into the full calculator.
Sources & review
Reviewed by CalcX Editorial Team
Updated 2026-10-07