Affinity Diagram

Use this skill to turn scattered qualitative inputs into clear, evidence-backed theme clusters.

Method Notes

• Affinity diagramming groups raw observations by natural relationship, not by a pre-decided framework.
• It is strongest when the input contains many notes, quotes, feedback items, ideas, symptoms, or concerns.
• The method should preserve traceability from each cluster back to the raw input.
• The output should help the next step: problem framing, prioritization, journey mapping, root-cause diagnosis, or solution design.

Required Inputs

Collect or infer these inputs before execution:

• raw items to cluster
• source or context of each item when available
• purpose of the clustering
• audience or decision context
• any known constraints, such as time, segment, process stage, or stakeholder group

If an input is missing, do not block automatically. Mark it as missing, state the assumption used, and add a validation action.

When Not To Use

Do not use when the categories are already fixed by a required taxonomy, compliance rule, financial chart of accounts, or formal decision criteria. Use mece-framework, decision-matrix, or weighted-scorecard instead.

Step-by-Step Execution

Output Template

### 1. Synthesis Question
- Purpose:
- Input scope:
- Decision or next use:

### 2. Affinity Map
| Cluster | Shared meaning | Raw items included | Evidence strength | Implication |
|---|---|---|---|---|
|  |  |  | strong / medium / weak |  |

### 3. Cluster Quality Check
| Cluster | Internally coherent? | Distinct from others? | Needs split / merge / keep |
|---|---|---|---|
|  | yes / partly / no | yes / partly / no |  |

### 4. Outliers And Gaps
- Outlier item:
- Why it may matter:
- Missing input:

### 5. Next Use
- Best next method or decision step:
- What to validate before acting:

Quality Gate

• Do not force raw items into a pre-existing framework unless the user requires one.
• Keep cluster labels specific; avoid vague labels such as "process issues" if the shared meaning is narrower.
• Preserve outliers instead of hiding them in miscellaneous.
• Avoid double-counting the same raw item across clusters unless it genuinely has two meanings.
• Keep traceability from conclusion back to source item.