What to remember
- Define options, criteria and scales before scoring.
- Keep observed performance separate from stakeholder preference.
- Remove overlapping criteria and enforce non-compensable constraints.
- Test sensitivity before writing the recommendation.
When is a multi-criteria matrix appropriate?
The method is useful when several comparable options pursue conflicting objectives and no single measure captures the decision. Market choices, initiatives or commitment scenarios may fit when they share the same scope and horizon.
Do not build a matrix when one option dominates across every relevant dimension, including separately assessed costs and constraints, when one mandatory threshold settles the question or when alternatives are not comparable. In those cases, scoring adds arithmetic without improving the decision.
| Situation | Useful treatment | Alternative |
|---|---|---|
| Conflicting objectives | Make trade-offs explicit | Multi-criteria analysis |
| Non-negotiable condition | Filter before scoring | Veto or eligibility rule |
| Monetisable costs and benefits | Analyse separately | Cost-benefit analysis |
| One option dominates every dimension | Check costs and constraints too | Direct documented decision |
Define criteria without counting the same value twice
Each criterion represents a distinct objective and has a definition, direction, scale and evidence source. Near-duplicate criteria give one benefit unintended extra influence. A mandatory constraint should not become a low-weight criterion: an option that fails it is screened out.
Use observable evidence for option performance whenever possible. When expert judgement is needed, retain the author, rationale and confidence level in the decision file.
- Criterion definition and objective
- Documented unit or scale
- Preferred direction
- Attributed evidence or judgement
- Missing-data rule
- Overlap check
Keep performance, component values, scales and weights separate
Raw performance comes from evidence in its own unit. A value function then maps it to a component preference value, such as 0 for the least-preferred admissible performance and 100 for the most preferred within the decision scope. A weight expresses the importance of the full swing between those anchors.
In a simple additive model, overall value is the sum of component values multiplied by weights that total 1. This requires coherent scales, accepted compensation and mutual preferential independence: the trade-off between two criteria must not depend on the level of a third. If it does, restructure the model or use a non-additive method. The result is neither economic value nor success probability.
| Register field | Question | Control |
|---|---|---|
| Raw performance | What does the evidence show? | Same unit per criterion across options |
| Component value | How desirable is that performance? | Explicit value function and 0-100 anchors |
| Weight | Which full performance swing matters most? | Normalised and attributed |
| Constraint | What cannot be compensated? | Screen before aggregation |
Worked example: compare three options with a weighted sum
Illustrative example: raw performances have already been mapped to component preference values, where 0 is the least-preferred admissible performance and 100 the most preferred within the decision scope. The 35%, 25%, 25% and 15% weights are an illustrative governance choice attributed to the decision-makers and apply to the full swing between those anchors.
| Option | Fit 35% | Evidence 25% | Reversibility 25% | Feasibility 15% | Score |
|---|---|---|---|---|---|
| A | 80 | 70 | 60 | 75 | 71.75 |
| B | 65 | 90 | 95 | 85 | 81.75 |
| C | 90 | 50 | 40 | 55 | 62.25 |
Test sensitivity and ranking reversals
B ranks first initially. The 70% fit and 10% for each other criterion weighting is an illustrative stress test, not an asserted plausible preference. It gives C 77.5, A 76.5 and B 72.5.
When only the fit weight varies and the others are redistributed proportionally, A ties B at 61%, C ties B near 63.48%, and C ties A at 66.67% before ranking first above that point. Distinguish stress tests from plausible weights elicited from decision-makers, and retain rank changes and evidence needs.
- Central weight set
- Documented plausible alternatives
- Ranks and score gaps
- Reversal criterion
- Evidence or test to obtain
- Review owner and date
Move from the score to a decision file
The conclusion is not “the highest score wins”. State the proposed option, benefits, weaknesses, verified constraints, weighting disagreements and the conditions that would reopen the decision.
The proposed editorial link within Innovatio associates Atlas with the decision context and record, and Grid with options, criteria and scores. It is neither a claim of publicly available capability nor proof of a deployed Atlas-to-Grid chain. The final trade-off remains human.
Frequently asked questions
Does the highest score automatically identify the best option?
No. It reflects the evidence, scales, compensation rules and stated preferences. Constraints, costs, risks and sensitivity remain part of the decision.
Must criteria always be weighted?
No. A non-aggregated trade-off table can be more faithful when criteria cannot compensate or defensible preferences are unavailable.
How should a qualitative criterion be handled?
Define observable levels with anchored examples, then retain the evidence or judgement supporting each assessment.
Is this method AHP, TOPSIS or ELECTRE?
Not necessarily. This guide describes a multi-criteria structure and a simple additive example. Specialised methods require their own assumptions and expertise.
Full sources
- An Introductory Guide to Multi-Criteria Decision AnalysisFull institutional guidance · Read the complete source
- Use of MCDA in options appraisalFull 14-page PDF · Read the complete source
- Multi-criteria analysis: a manualFull 168-page manual · Read the complete source
- Multi-criteria assessment of innovation projects under uncertaintyFull open-access research article · Read the complete source