Decision matrix
Compare eligible options against shared criteria and make the effects of your weights and ratings visible.
This template uses weights totaling 100% and ratings from 1 to 5. A higher rating must always mean a better outcome, including for cost criteria.
Hard constraints, at least two eligible options, weighted criteria, complete ratings, and the evidence behind them.
Calculated weighted scores alongside the original inputs, assumptions, and trade-offs.
How to use it
- Define hard constraints and keep only eligible options.
- Agree on criteria, weights, and a consistent rating scale.
- Enter evidence-based ratings in the same criterion order.
- Compare the calculated scores and test close results with plausible input changes.
Keep its limits in mind
Scores reflect the inputs; they do not prove an option is best. Screen out options that fail hard constraints and test whether small input changes alter close results.
Comparing three member-management tools
Fictional example: Qinghe compares three member-management tools, all eligible under a CNY 20,000 annual cap and the required permissions. The teaching ratings show a close result, not an approved choice.
On a small screen, scroll across the diagram to read it.
Close result: The 0.05 gap between A and B is small. Different supported ratings or priorities may reverse their order.
Calculation: Each score is the sum of rating multiplied by weight, divided by 100. The application recalculates it when inputs change.
A complete sensitivity scenario: Change all weights together to 50/30/20, totaling 100%. A scores 4.10, B 4.20 and C 3.70. The initial 40/35/25 scores are 4.10, 4.05 and 3.85. B leads in the alternative scenario; neither ranking grants approval.
Next step: Check rating evidence and compare plausible weight changes before committing.
Three reasons to postpone scoring
Use a decision matrix to expose tradeoffs among feasible alternatives. It is most useful when participants agree on the decision boundary but differ about priorities. Before opening the rating grid, write the decision, options, non-negotiable requirements and decision owner. Separate an annual purchase limit from total implementation effort; satisfying one does not establish the other.
Postpone scoring when eligibility is unresolved. An option without a required permission cannot offset that failure with excellent usability. Postpone when the criteria repeat the same benefit under different names, such as “easy to learn” and “short training time,” unless their different meanings are explicit. Postpone when no one can explain what a rating means. Giving an unknown a middle score creates information that the team does not possess.
Define each scale before rating the named options. A higher affordability rating must mean more favorable cost, not more expensive. Describe which costs and period are included, and keep omitted migration or support costs visible. For subjective criteria, record the judgment and its author rather than disguising it as a measured result. Disagreement can remain in the evidence column until the group decides whether another observation is worth obtaining.
Choose a comparison that matches the evidence
A weighted grid is only one way to compare choices. If numbers would hide the central issue, a smaller qualitative comparison may be more useful. Keep the hard-constraint screen outside the weighted total in every format.
| Method | Use when | What it does not establish |
|---|---|---|
| Decision matrix | Eligible options have shared criteria and defensible scales | The highest score is not approval or a success probability |
| Pros and cons list | Important tradeoffs are difficult to rate honestly | A longer list is not automatically a better option |
| TOWS | You still need action candidates from strategic factors | A factor combination is not a ready-to-buy alternative |
| SMART | A proposed option lacks a checkable target | A measurable target is not evidence the option can achieve it |
Worked continuation: a 0.05 lead reverses
Use the fictional member-management tool case above. The teaching packet stipulates that A, B and C meet the CNY 20,000 annual cap and required permissions. That is an exercise assumption, not a current vendor quotation or legal assessment. The supplied ratings are also practice judgments: customer value, ease of use and affordability, each from 1 to 5 with higher meaning more favorable. No real customer test results are supplied.
The initial weights are 40, 35 and 25, totaling 100. Calculate A as (4×40 + 5×35 + 3×25)/100 = 4.10. B is (5×40 + 3×35 + 4×25)/100 = 4.05. C is (3×40 + 4×35 + 5×25)/100 = 3.85. A leads B by 0.05 on this agreed scale. It does not have a 5% greater chance of success, nor is a 4.10 score evidence of a financial return.
Now change the complete weight set to 50, 30 and 20, leaving ratings fixed. The weights still total 100. This is a sensitivity scenario, not a claim that the decision owner has accepted new priorities. B now leads A by 0.10. Keep both scenarios so a reviewer can see exactly which assumption changed, instead of replacing the first table with a preferred answer.
| Option | Value / ease / affordability | 40/35/25 score | 50/30/20 score |
|---|---|---|---|
| A | 4 / 5 / 3 | 4.10 | 4.10 |
| B | 5 / 3 / 4 | 4.05 | 4.20 |
| C | 3 / 4 / 5 | 3.85 | 3.70 |
Make the request proportionate to the evidence
A useful recommendation is: “The initial priorities narrowly favor A, but the alternative priorities favor B. Before choosing, confirm how much emphasis belongs on customer value and inspect the rating evidence.” This preserves the actual tradeoff. It is stronger than claiming that the model selected A, because it identifies what could change the answer and what information is missing.
Prepare a rating ledger with the criterion definition, source, observation period, evaluator, uncertainty and reasons for the number. In this practice case those supporting observations are absent. In real procurement, request comparable demonstrations or records before treating the grid as a buying recommendation. A sensitivity check on weights does not repair unsupported ratings. If one rating remains unknown, keep it empty and explain why a complete score is not yet available.
For the decision meeting, retain the original values, the alternative set, hard-constraint evidence and the precise request: further evaluation, a limited trial or a purchase. Record who actually approved which scope and conditions. Later review implementation effort, user outcomes and the assumptions that changed; do not use a successful installation as proof that customer value improved. Reopen the eligibility screen if permissions, costs or service terms change before commitment.
Check your work
- Options meet the hard constraints.
- Weights total 100%.
- Every option has a rating for each criterion.
- Higher ratings consistently mean better outcomes.
- Evidence and sensitivity are reviewed before a decision.
Framework references
References explain the framework and its scope; they do not validate the fictional example's results.
Put the framework to work
Copy the outline and add your question, evidence, and open questions.
Read a practical guide →Practice with an example
Recalculate a fictional procurement ranking reversal, check budget limits and comparable scores, and record tradeoffs and unresolved conditions.
When a weighted decision matrix changes its winner →