Affinity Cluster

Origin & evolution

The technique comes from the Japanese anthropologist Jiro Kawakita, who developed it in the 1960s and gave it his initials, the KJ method. Kawakita built it to make sense of fieldwork, organising qualitative findings from ethnographic studies, including work in Nepal on environmental equilibrium, where conventional quantitative methods left the data inert. His insight🔦InsightUser ResearchA synthesised finding from researchView reference → was procedural: write each observation on its own card, then group the cards by felt relatedness before naming any group, so the structure follows the data instead of the researcher's prior categories.

The method spread well beyond anthropology. It became one of the Seven Management and Planning Tools of Japanese quality management, where it is known as the affinity diagram, and from there entered software and UX research as affinity mapping. The wall of sticky notes that teams cluster after a round of interviews is the KJ method, usually practised without anyone knowing whose method it is.

A discipline underlies the ritual that is easy to skip. The grouping is meant to be silent and intuitive first, with labels written only after the clusters settle, which is why the method leans on what the philosopher Charles Peirce called abduction, reasoning toward the best explanation, over deductive sorting into known buckets. Two failure modes recur: clustering by surface topic, such as putting every note that mentions "search" together regardless of what each reveals, and naming groups so early that the labels dictate the sort. Done well, the cluster surprises the team that made it.

How it works in practice

After fifteen interviews about a budgeting app, a team has roughly ninety observations on cards. They sort in silence. One cluster forms from notes that mention no shared keyword: "checks balance before buying coffee," "screenshots the total to send a partner," "feels guilty opening the app after a big weekend." Topically these have nothing in common, yet they group by an emotional thread, money as a source of shared anxiety rather than a number to optimise. The team names the cluster only once it has settled: "the app is read as a verdict, not a tool." That label becomes a single synthesised insight, and the ninety cards behind it stay available as the evidence🔎EvidenceValidationData supporting or refuting a hypothesisView reference → trail. A topic-first sort would have scattered those three notes across "balance," "sharing," and "notifications," and the theme would never have appeared.

Affinity cluster vs. its neighbours

• Observation is the single card, one thing seen or heard. The cluster is the emergent grouping of many. Clusters have no meaning without the observations underneath, which is why a cluster should always be able to list its members.
• Insight is the named conclusion a settled cluster yields, the durable, decision⚖️DecisionStrategyA recorded decision with context, rationale, and consequencesView reference →-shaping statement. The cluster is the working structure that produces it; the insight is the output. One good cluster typically resolves into one insight.
• Research study🧑‍🔬Research StudyUser ResearchA planned research activityView reference → is the session or programme that generated the raw notes. The cluster is an analysis artefact built from a study's output, the bridge between collecting data and concluding from it.

In the graph

In the Unified Product Graph, an affinity cluster is the synthesis structure of the User Research region, sitting deliberately between raw observations and finished insight. A body of fieldwork resolves into clusters through research_study_clusters_into_affinity_clusterResearch Studyclusters intoAffinity Clusterhierarchy, which records that a set of notes was grouped rather than cherry-picked. Each settled cluster then yields its conclusion through affinity_cluster_synthesises_insightAffinity ClustersynthesisesInsighthierarchy, making the analytical leap an explicit, inspectable edge. That structure preserves the audit trail the KJ method was built to honour: every insight traces back through its cluster to the individual observations that justified it, so a stakeholder🧑‍💼StakeholderTeam & OrganisationA person with influence over the productView reference → can always ask which notes produced a given theme and get a real answer.