
A decision matrix template is a table for comparing a short list of options against the same criteria. In the weighted version, each criterion gets a weight, and the weights add up to 100%. You score every option from 1 to 5 on each criterion and multiply each score by its weight. Add those products and divide by 100 to get a total on the same 1–5 scale.
The total supports a decision. It does not make it. It is worth taking seriously only after two more checks: whether a reasonable change to the weights would change the winner, and whether a written record explains why the team chose what it chose.
This page gives you both. It has a free template in three formats and a fully worked example in which every score and calculation is shown. It also has a sensitivity check that finds the exact weight at which the winner changes, and a decision record you can copy.
Free downloads (no sign-up):
A decision matrix works best when you already have two to five realistic options and several factors that pull in different directions. It also helps when different people value those factors differently. Writing the criteria and weights down turns a vague argument ("Kanban feels better") into a specific one ("you weighted predictability at 20%; I think it is 30%").
It is the wrong tool in other situations:
| Situation | Better approach |
|---|---|
| You do not have options yet, only a pile of ideas | Cluster the ideas first with an affinity diagram, then shortlist |
| One factor decides it (a legal requirement, a hard budget cap) | Treat that factor as a pass/fail gate and pick among whatever passes |
| The choice is cheap and easy to reverse | Try one option for a short period; a matrix costs more than it saves |
| The choice is a chain of if/then outcomes with probabilities | Use a decision tree |
| You are ranking tasks, not choosing between alternatives | Use an Eisenhower matrix or an impact–effort grid |
| Criteria interact (A is only good if B is also high) | Adding up weighted scores assumes the criteria are independent, so discuss the combinations directly |
| The decision is mainly an ethical or values judgment | Use the matrix to structure the discussion, not to produce the answer |
Government decision-analysis guidance makes the same point. The UK Government Analysis Function describes multi-criteria analysis as a tool that helps decision-makers, not a complete way of making the decision (Analysis Function, 2024).
The template has six parts. Fill them in order, because each one constrains the next.
1. Gates (must-haves). Mark each option pass or fail. An option that fails any gate is removed and gets no score.
| Gate (must-have) | Option A | Option B | Option C |
|---|---|---|---|
| G1: | Pass / Fail | Pass / Fail | Pass / Fail |
| G2: | Pass / Fail | Pass / Fail | Pass / Fail |
2–6. Criteria, anchors, weights, scores and totals.
| Criterion | What a 5 and a 1 look like | Weight (%) | Evidence | Option A | Option B | Option C |
|---|---|---|---|---|---|---|
| C1 | 5 = … / 1 = … | |||||
| C2 | ||||||
| C3 | ||||||
| C4 | ||||||
| C5 | ||||||
| Total | 100 |
Each option's total is Σ(weight × score) ÷ 100.
Scoring scale. Use the same scale for every criterion and write the anchors before anyone scores.
| Score | Meaning |
|---|---|
| 5 | Exceeds the requirement, with evidence you could show someone |
| 4 | Meets the requirement well; minor gaps |
| 3 | Meets the minimum you wrote down |
| 2 | Partly meets it; needs a workaround |
| 1 | Does not meet it |
A 1–5 scale is usually enough. Wider scales such as 1–10 invite debates about whether something is a 6 or a 7, without the evidence to settle them.
For "lower is better" criteria such as cost or effort, flip the wording so that 5 is always good. For example, write "Adoption effort (higher = easier)". That way, a higher total always means a better option.
A simple (unweighted) decision matrix adds up the scores, so every criterion counts equally. A weighted decision matrix multiplies each score by how much that criterion matters before adding up.
| Simple matrix | Weighted matrix | |
|---|---|---|
| Calculation | Sum of scores | Σ(weight × score) ÷ 100 |
| Assumes | Every criterion matters equally | Some criteria matter more, and the team agrees by how much |
| Best for | Quick first pass, low-stakes choices | Choices where priorities genuinely differ |
| Main risk | Minor criteria can outvote the critical one | Weights can be tuned to produce a preferred winner |
The example below shows why the difference matters. With no weights, two-week sprints and Kanban tie at 20 points each. Weighting by the team's agreed priorities breaks that tie.
Converting relative weights to percentages. Many templates use weights like 3, 2 and 1. To convert them, divide each by their sum and multiply by 100. With 3 + 2 + 1 = 6, the weights become 50%, 33.3% and 16.7%.
The ranking does not change, but percentages are easier to audit. Everyone can see that the first criterion carries half the decision, and the sensitivity check later on is easier to read.
This example is fictional. The team, scores and outcome were made up to teach the method. They are not a claim that one workflow is better than another in general, and they do not describe a real AFFiNE customer.
The decision. A 12-person product team must decide how to run project work for the next two quarters. Three to five customer escalations a week pull engineers off planned work. Leadership wants dated estimates for two milestones that carry external commitments.
The options. The team shortlisted four:
| Gate | Sprints | Kanban | Cycles | Fixed plan |
|---|---|---|---|---|
| G1: Scope can change at least once a quarter without steering-group re-approval | Pass | Pass | Pass | Fail |
| G2: Runs with current people and tools (no new hires or paid software before Q2) | Pass | Pass | Pass | Pass |
The fixed-scope plan fails G1 and leaves the comparison. However well it might score on predictability, it cannot buy back a requirement the team has declared non-negotiable.
| # | Criterion | What a 5 and a 1 look like | Weight |
|---|---|---|---|
| C1 | Responsiveness to new priorities | 5 = a new top priority can start within 2 days; 3 = within 2 weeks; 1 = waits more than 6 weeks | 25% |
| C2 | Forecastable delivery dates | 5 = can give dated ranges for quarterly milestones from the team's own history; 1 = no basis for a date | 20% |
| C3 | Protection from interruptions | 5 = weekly escalations are absorbed without breaking planned work; 1 = every escalation breaks the plan | 20% |
| C4 | Adoption effort (higher = easier) | 5 = no new roles or meetings; 3 = a few new routines one person has run before; 1 = new roles and skills nobody has | 15% |
| C5 | Stakeholder visibility | 5 = stakeholders see working results every 2 weeks or more often; 1 = only on request | 10% |
| C6 | Documentation and handoff fit | 5 = the method itself produces written scope and decisions; 1 = relies on memory and chat | 10% |
| Total | 100% |
Each cell shows the 1–5 score and its weighted contribution (weight × score ÷ 100).
| Criterion | Weight | Two-week sprints | Kanban with WIP limits | Six-week cycles |
|---|---|---|---|---|
| Responsiveness | 25% | 3 → 0.75 | 4 → 1.00 | 2 → 0.50 |
| Forecastable dates | 20% | 4 → 0.80 | 2 → 0.40 | 4 → 0.80 |
| Interruptions | 20% | 2 → 0.40 | 4 → 0.80 | 3 → 0.60 |
| Adoption effort | 15% | 3 → 0.45 | 4 → 0.60 | 2 → 0.30 |
| Visibility | 10% | 5 → 0.50 | 3 → 0.30 | 4 → 0.40 |
| Documentation | 10% | 3 → 0.30 | 3 → 0.30 | 4 → 0.40 |
| Weighted total | 100% | 3.20 | 3.40 | 3.00 |
| Simple sum (no weights) | 20 | 20 | 19 |
The arithmetic in full:
Provisional result: Kanban with WIP limits leads with 3.40, ahead of two-week sprints at 3.20 and six-week cycles at 3.00. A 0.20 margin on a 1–5 scale is small. The next step is to check whether it holds up.
A sensitivity check asks one question: if we had weighted things a little differently, would we pick a different option? The UK multi-criteria analysis manual includes it among the steps of a multi-criteria analysis, testing the results against changes in scores or weights (DCLG, 2009). NASA's systems engineering handbook likewise asks analysts to tell the decision-maker how robust a ranking is (NASA SP-2016-6105 Rev2).
Test the weight people argued about. Here, the sales lead argued that dated forecasts deserved more than 20%. To keep the total at 100%, raise predictability by 10 points and take those 10 points from responsiveness. The other four weights stay the same.
| Baseline | Test | |
|---|---|---|
| Responsiveness weight | 25% | 15% |
| Predictability weight | 20% | 30% |
| Two-week sprints | 3.20 | 3.30 |
| Kanban with WIP limits | 3.40 | 3.20 |
| Six-week cycles | 3.00 | 3.20 |
Recalculated with the same scores:
The winner is not robust to this change. At 30% predictability, sprints lead and Kanban drops into a tie with cycles.
A single test tells you that the winner changes. The break-even weight tells you where, which is the number your stakeholders actually need to discuss.
Moving x points from responsiveness to predictability changes each option's total by x × (predictability score − responsiveness score) ÷ 100:
So Kanban stays ahead while the predictability weight is below about 26.7%. If weights are whole percentages, sprints take the lead at 27%.
The workbook's sensitivity table does this sweep for you. Choose which criterion to raise and which to lower, and it recalculates every option at shifts from 0 to 20 points, showing the leader at each step.
How to read the result. The matrix has turned "which workflow?" into a narrower question: is predictability worth more or less than about 27% of this decision? That question can be answered with facts.
In the example, the head of product confirmed that only two milestones carry external dates. The rest of the roadmap can move. On that basis, the team kept predictability at 20% and wrote the 27% threshold into the decision record as a review trigger.
Anything you would never trade away belongs in a gate, not in the weights. A weighted total is compensatory: a high score on one criterion can make up for a low score on another. Mandatory requirements should not work that way.
The UK multi-criteria analysis manual makes the same point: where options must reach a minimum acceptable level, those below it are rejected outright, because better performance elsewhere cannot compensate. It also warns that thresholds chosen after seeing how options score can bias the result, so set gates before scoring (DCLG, 2009). Keep gates few and genuinely binary, such as a legal requirement, a hard budget ceiling or a security control. A gate that most options fail is often a preference written too strictly.
Decide your tie rule before you see the totals. A workable order:
Do not fill a gap with a guess presented as a fact. Leave the cell blank and mark it, which is what the workbook's "Missing scores" column counts. Or score it conservatively and set the confidence to Low.
Then ask whether the gap could change the decision. This is a quick value-of-information test.
Most bias in a decision matrix enters through the weights and through group scoring.
A matrix shows how the options scored. A decision record shows what was decided, why, who disagreed and when the question should be reopened. Without one, six months later nobody remembers whether 20% predictability was a deliberate choice or an oversight.
| Field | What to write |
|---|---|
| Decision | What was decided, in one sentence |
| Owner and date | Who is accountable, and when |
| Options considered | Every option, including any screened out and why |
| Gates applied | Pass/fail requirements checked before scoring |
| Criteria and weights | Matrix version, and who agreed the weights before scoring |
| Result | Weighted totals and the margin between first and second |
| Sensitivity finding | Which weight change would flip the result, and at what point |
| Missing or weak evidence | Low-confidence scores, and whether they could change the result |
| Why this option | The reasoning in plain words, not just the score |
| Dissent | Who disagreed, about what |
| Mitigations | What will be done about the chosen option's weak scores |
| Review trigger | The event or date that reopens the decision |
In the example, the record says:
Mitigations do not change the scores already given. If you want to score "Kanban plus fortnightly demos" as a different option, add it as a new column and rescore it.
Store the record next to the matrix and the evidence it cites. The same principle runs through good project documentation practices: decisions should be findable by people who were not in the room. If the decision belongs to a project's planning stage, file it with the other planning-phase deliverables and link it from the project plan. Track the open risks it creates in a RAID log.
These tools are often confused because several are called "matrices". They answer different questions.
| Tool | Question it answers | Typical input |
|---|---|---|
| Weighted decision matrix | Which of these options should we choose? | 2–5 alternatives × several weighted criteria |
| Impact–effort matrix | Which tasks or ideas should we do first? | A backlog plotted on two axes, impact and effort |
| Eisenhower matrix | What do I do now, schedule, delegate or drop? | Tasks sorted by urgency and importance |
| Pugh matrix | Is each concept better or worse than a baseline? | Concepts scored +/0/− against a reference design |
| Pros-and-cons list | What are the arguments for and against? | One or two options, unscored |
| RACI chart | Who does what in carrying out the decision? | Roles × tasks (try the RACI matrix template) |
ASQ lists "Pugh matrix" as another name for a decision matrix, but in practice a Pugh matrix usually rates concepts as better, same or worse than a baseline design, while a weighted matrix scores every option on an absolute scale.
When the options differ mainly in money over several years, a cost-benefit analysis template with NPV and payback is usually the better tool, and its result can feed one criterion of your matrix.
The most common mix-up is between the decision matrix and the impact–effort matrix. A decision matrix is for choosing one option from alternatives, using several criteria at once. An impact–effort matrix is for ordering many items by two dimensions only.
They work in sequence. Use the decision matrix to pick the workflow. Then sort the work that follows with an impact–effort matrix template or the Eisenhower matrix template. Where the choice is about which risks to accept rather than which option performs best, a risk assessment template is a better starting point.
Excel. Open the .xlsx file. White cells are inputs; shaded cells contain formulas.
The workbook has five sheets:
The Check cell warns when weights do not total 100%, when scores are missing, when there is a tie and when the result is close.
Google Sheets. Upload the .xlsx file to Google Drive and open it with Google Sheets. We checked the workbook by recalculating it in LibreOffice Calc. It uses only standard functions that Excel and Google Sheets also provide (SUMPRODUCT, RANK, INDEX, MATCH, COUNTIF, ROUND), but after opening it in another app, confirm that the "Worked example" sheet shows totals of 3.20, 3.40 and 3.00. The .csv files hold values only and do not recalculate.
AFFiNE. AFFiNE can import Markdown files (Import → Markdown files), and Markdown tables come in as table blocks. Import the .md template to get the blank matrix, sensitivity table and decision record as one editable doc. Each table's header arrives as an ordinary first row.
As of October 2026, AFFiNE tables and databases have no formula columns, so they cannot calculate weighted totals. Do the arithmetic in the spreadsheet and keep the decision in AFFiNE, where you can:
@ mentions, so the evidence sits next to the decision.
Accessibility note. If you share the matrix as an image or color-coded table, print the number in every cell, as the heatmap above does. Do not rely on color alone, and provide the table as text so screen-reader users get the same information.
Checked on 5 October 2026.
Last reviewed: October 5, 2026. The worked example and every number in it are fictional teaching data. The template files are generated from the same data so that the article, spreadsheet and images stay consistent.