To build a decision matrix with AI, define the choices, criteria, and scoring rules before asking for a ranking. Then calculate the totals and check whether a reasonable change in priorities changes the result.
A matrix makes tradeoffs visible. It cannot turn uncertain estimates into measured facts. This guide uses three fictional documentation projects with invented ratings for an arithmetic exercise. The scores are not research findings or claims about how these formats perform in general.
1. Define the decision and any firm requirements
Imagine a small team choosing its next help-content project: a screenshot tour, a task-based guide, or a short FAQ. The decision is which one to prepare first, not whether the other formats are useful.
List firm requirements before scoring. For example, if the work must fit an approved schedule, a choice that cannot meet that schedule should be flagged as ineligible rather than allowed to compensate with a high score elsewhere. In our fictional exercise, assume all three choices pass the team’s initial eligibility check.
ASQ’s decision-matrix guidance describes comparing options against weighted criteria. It also notes the need to keep scoring direction clear for criteria such as cost or difficulty. We use “ease” below so that a higher number consistently means a more desirable result.
2. Make the scoring direction explicit
Use a five-point scale for each criterion. Define what the endpoints mean before scoring options:
- Usefulness for the chosen user task: 1 means little expected help; 5 means strong expected help.
- Delivery ease for this team: 1 means difficult to produce with current resources; 5 means easy to produce.
- Upkeep ease: 1 means difficult to keep current; 5 means easy to maintain.
The initial weights are 40% for usefulness, 35% for delivery ease, and 25% for upkeep ease. They sum to 100%. A weight expresses the team’s priority; a rating expresses how the team assesses an option against one criterion. Keep those two decisions separate.
Avoid scoring the same preference repeatedly under similar names. If “delivery ease,” “low effort,” and “speed to publish” all reflect the same estimate, giving each its own weight can give that preference more influence than intended.
3. Record the ratings and their assumptions
| Option | Usefulness | Delivery ease | Upkeep ease |
|---|---|---|---|
| A: Screenshot tour | 3 | 5 | 2 |
| B: Task guide | 5 | 2 | 4 |
| C: Short FAQ | 2 | 4 | 5 |
For this fictional team, the screenshot tour is assumed to be quick to assemble from existing material but harder to maintain as screens change. The task guide is assumed to address the selected user task well but require more drafting. The FAQ is assumed to be easy to maintain but less helpful for that particular task.
These assumptions explain the sample ratings; they do not establish universal properties of tours, guides, or FAQs. In a real decision, add evidence beside each rating, such as a small user test or an estimate from the person doing the work. Mark unsupported ratings as provisional instead of asking AI to fill every cell with confident numbers.
4. Ask AI to calculate and expose assumptions
Build a weighted decision matrix using only these supplied ratings. All options have passed our initial eligibility check. Higher scores are better for every criterion. Criteria, in order: 1. Usefulness 2. Delivery ease 3. Upkeep ease Ratings: A, Screenshot tour: 3, 5, 2 B, Task guide: 5, 2, 4 C, Short FAQ: 2, 4, 5 Initial weights: 40%, 35%, 25%. Calculate each weighted total and show the arithmetic. Do not invent supporting evidence or change the ratings. Then recalculate with weights of 25%, 50%, 25%. Explain whether the ranking changes and retain any ties. Separate the numerical result from assumptions that need verification.
If your evidence begins as open-ended comments, use a separate process to organize it before assigning scores. Our customer-feedback guide explains why comment counts alone do not establish severity or priority.
5. Check the totals independently
Multiply each rating by its weight as a decimal, then add the three contributions. For the initial weights:
A: (3 × 0.40) + (5 × 0.35) + (2 × 0.25) = 3.45 B: (5 × 0.40) + (2 × 0.35) + (4 × 0.25) = 3.70 C: (2 × 0.40) + (4 × 0.35) + (5 × 0.25) = 3.45
B ranks first under these inputs; A and C tie. The 3.70 total is a weighted score on the chosen scale, not a probability of success or a measured percentage improvement. Two decimal places reflect the calculation, not certainty in the ratings.
Check the weight total, criterion order, and multiplication yourself. A correct formula applied to swapped columns still gives a wrong result for your intended comparison. If you put the calculation in a spreadsheet, our AI formula-checking guide covers using known answers to test it.
6. See what happens when priorities change
Now suppose delivery ease matters more. Change the weights to 25%, 50%, and 25%, keeping all ratings unchanged:
A: (3 × 0.25) + (5 × 0.50) + (2 × 0.25) = 3.75 B: (5 × 0.25) + (2 × 0.50) + (4 × 0.25) = 3.25 C: (2 × 0.25) + (4 × 0.50) + (5 × 0.25) = 3.75
A and C now share first place. That is a reason to discuss the priorities, not to ask AI to invent an extra decimal or an arbitrary tie-breaker. If the team still considers usefulness the main goal, the initial weighting may be appropriate. If resources have tightened, the alternative weighting may better describe the decision.
You can also test uncertainty in ratings. Under the initial weights, reducing B’s provisional usefulness rating from 5 to 4 lowers its total from 3.70 to 3.30, below A and C at 3.45. That one-point change shows why checking the usefulness assumption could be more valuable than polishing the ranking explanation.
7. Save a decision with its conditions
A useful decision note would say: “The task guide ranks first at 3.70 under weights of 40%, 35%, and 25%. This depends on provisional ratings, especially usefulness. When delivery ease receives 50% weight, the screenshot tour and FAQ tie for first.”
Keep the eligibility rules, ratings, evidence, weights, and alternative scenario together. Choose a next step that resolves an important uncertainty, such as testing a small guide section with intended readers or confirming the drafting estimate. AI can help organize the comparison; the recorded assumptions explain when its ranking should be reconsidered.
Put the decision into a realistic plan
After selecting an option, use the daily scheduling workflow to fit the next steps around fixed events. A high score does not create extra working time; check task durations and decide what to postpone if the work runs over.
For another task, browse the AI guides by workflow.