Affinity Map
The affinity map is a collaborative game of sense-making. Participants surface a large number of ideas, observations or requirements on notes, then move those notes on a wall or a board to bring the similar ones together, name the groups that emerge and read the structure that appears. The approach is bottom-up: no category is decided in advance, the grouping is discovered from the material itself. The result is a small set of named clusters that turns a mass of scattered entries into a readable thematic structure a team can act on.
Goal
The affinity map turns dozens of raw, unordered entries into a handful of named themes, so the group sees the shape of the problem. Those entries come out of a brainstorming session, a run of interviews, a retrospective or an open-ended survey. It belongs to the family of collaborative games: within that family it handles the synthesis question, where the Product Box handles the vision of an offering and the Fishbowl the perspectives of two groups that do not listen to each other in an ordinary discussion.
The technique supports a decision: which themes stand out, which recur most and therefore where to steer the analysis, the requirements work or the improvement effort next. The density of a cluster, the number of notes it gathers, is the telling signal: it flags the consensus or the prevalence of a theme in the harvested material.
The deliverable is a grouped map: notes arranged under headings born of the sort, plus a parking lot for the outliers. The named clusters and their relative sizes are the artefact that feeds the next step. The sorting is done by the group, because the value sought is a shared model: building the clusters together creates buy-in and surfaces the hidden assumptions and the interpretive divergences that a desk analysis would never see. An analyst who sorts the cards alone has done a filing job, whereas the game earns its worth through the joint making.
Usage
When to use it
- A large volume of unstructured input: twenty items or more to order, workshop notes, free-text answers, retrospective cards.
- An unknown structure to bring out: no preset categories, you want it to emerge from the material.
- Stakeholders with different mental models: you are after a negotiated grouping and the buy-in its joint construction produces.
- Upstream elicitation or framing: consolidating needs, organising requirements before prioritising them.
- An interpretive disagreement to expose: the shared sort surfaces the assumptions each group takes as obvious.
When not to use it
- Categories already known or a fixed taxonomy: you are filing into predefined boxes, a closed card sort does the job at lower cost.
- Few items, all well understood: no hidden structure to reveal, settle it with a facilitated discussion or a ranking.
- A goal of ranking or explaining a cause: the map organises without prioritising or showing causality, prefer the fishbone diagram for the cause.
Description
Collaborative games share a common frame, three timed phases and a neutral facilitator. The affinity map works it around a single operation: moving notes closer together until a structure appears. The method goes back to the Japanese ethnographer Jiro Kawakita, who formalised it in the 1960s under the name KJ method, after his initials, from an observation: faced with masses of field data, laying the cards out in space brought out a meaning that linear reading left hidden. The affinity diagram now sits among the seven quality management tools, the door through which the world of quality and business analysis took it up.
The elements of a session
The session rests on five elements. A generative question, a clear open statement able to prompt at least twenty answers; the whole exercise is worth only what that question is worth. Notes, one idea per note, short and legible. A large shared surface, wall, whiteboard or digital board, where everyone sees the whole at a glance. A neutral facilitator, who holds the rules, keeps the pace and draws everyone in, without ever imposing the grouping. And a parking lot, a corner set aside for the notes that fit nowhere, so that nothing is forced into a cluster.
How the session runs
The question is asked and shown visibly. Each participant then writes their notes in silence, around ten minutes, one idea per note. Silence is a discipline: it prevents early anchoring, the moment when the first idea spoken steers all the rest, and it gives the more reserved the same weight as the dominant voices, an effect the BABOK notes when it observes that the game pushes usually quiet participants into an active role. Discussing each note as it is written reintroduces exactly the anchoring that silence avoids. Below about twenty notes there is nothing to discover and the map merely repeats its inputs: the raw material has to be abundant enough.
All the notes are then displayed, and the group moves them to bring the similar ones together. Two disciplines govern this sort. It is led by the participants. A facilitator who sorts in their place produces their own model and destroys the buy-in that is the whole point of the game: that is the cardinal mistake of the technique. And it ideally starts in silence, one note at a time, so that no voice dictates the arrangement before the material has settled, the discipline proper to the KJ method. Group first, name later: deciding a category before the notes have stabilised biases the sort toward that label. Duplicates stay visible, because two identical notes signal a consensus. Outliers go to the parking lot rather than being forced into the nearest cluster, a forcing that tidies the wall but loses the signal the note carried.
Once the groupings are stable, the group proposes a heading for each column. The name is a working label: if "Facilities" and "Infrastructure" split the room, you write both or keep the most approved label and move on, rather than debating a word for fifteen minutes. The number of clusters is watched: too many and the pattern dissolves into noise; too few and everything collapses into a catch-all "Miscellaneous". A legible handful, four to eight most often, is the right order of magnitude.
The group finally reads the resulting structure: which themes are the fullest, what surprises, what the relations between clusters imply and it decides the next action, prioritise, assign, pour into the requirements. The map organises and makes things comparable. Ranking the themes or tracing their causes belongs to distinct steps: the map comes before a prioritisation and prepares it. A second sorting pass is sometimes useful to find deeper or cross-cutting affinities.
AI considerations
AI helps upstream of the human sort, on what will not fit on a wall. A language model or a word-embedding model groups hundreds of free-text survey answers, support tickets, reviews or interview transcripts into candidate themes, a volume the group would never sort by hand. It also proposes cluster headings and spots the near-identical notes to merge. In a multilingual Swiss setting, it normalises notes written in French, German and Italian so that a mixed team sorts a single shared set.
The limit is that shared meaning is the very value of the technique. A model that returns finished clusters short-circuits the collective negotiation that builds understanding and buy-in: AI serves to produce a draft the group reworks. It groups by lexical or semantic proximity, whereas humans group by intent, context and meaning: two notes that state the same need in different words or the same word for two distinct needs are commonly misplaced by the machine and correctly separated by those who wrote them.
Data sensitivity weighs in at the end. The notes often carry confidential stakeholder feedback or personal data, and handing them to an external service engages the Swiss data-protection duties under the revised Federal Act on Data Protection (revFADP). Sensitive material stays on an internal or approved tool; failing that, it is anonymised before any automated pass.
Examples
A requirements-discovery workshop in a health insurer consolidates what the insured expect from a new self-service portal. Twelve participants produced about fifteen notes, sorted into named clusters.
Affinity Map
From scattered notes to themes that emerge
The sort brings out four themes no one had set in advance. The fullest cluster, "Statements and reimbursements", concentrates the most notes: its density flags where the insured feel the most friction and steers the analysis effort toward it. The headings, "Deductible and insurance model", "Proactive communication", "Mobile access and identity", were born of the grouping: the material dictated the categories. The detail not to miss is the parking lot, set apart: "health chat assistant" and "loyalty programme" fit no cluster and stay visible there rather than forced into the nearest box, because an outlier often carries a signal, a fresh idea or a marginal need, that would be lost by filing it away. The map makes the themes visible and comparable; ranking them belongs to the next step.
Visualisations
The affinity map is a spatial artefact: the position of the notes carries the meaning, and three visual signals read off the resulting map.
| Spatial signal | What it says |
|---|---|
| Neighbouring notes | Same theme, an affinity felt by the group. |
| Height of a column | Prevalence and consensus: the fuller a cluster, the more shared the theme. |
| Note in the parking lot, set apart | An outlier kept, a potential signal rather than noise to discard. |
Cost
| Phase | Level | Rationale |
|---|---|---|
| Preparation | Low to medium | A good generative question, the material (wall and notes or a digital board) and the invitation. The question is the real preparatory work. |
| Execution | Low | A single session; the format fits in about an hour and a half for a group of up to twenty. |
| Documentation | Low to medium | Photograph or export the board, then transcribe the clusters and the note counts. Light on a digital board that captures on its own, heavier from a physical wall. |
Tooling
The original format is tactile: a wall or a whiteboard, repositionable notes, markers. It suits in-person sessions, where the physical act of moving a note keeps up the energy and engagement; its downside is manual documentation and an ephemeral artefact. Digital whiteboards, Miro, Mural, FigJam, Conceptboard, Microsoft Whiteboard, suit remote or hybrid teams and document themselves, the board being its own exportable archive. For a Swiss organisation, data residence steers the choice when sensitive feedback has to stay in the region: Conceptboard hosts in Europe and Microsoft Whiteboard already lives in the M365 suite that many Swiss organisations run. For very large text corpora, a spreadsheet coupled with AI-assisted or NLP grouping makes a first pass, producing draft clusters the group then re-sorts on a board: the tool scales the input, the group keeps the sense-making.
Sources
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.10 Collaborative Games.
- Scupin, R., "The KJ Method: A Technique for Analyzing Data Derived from Japanese Ethnology", Human Organization 56(2), 1997 (Society for Applied Anthropology): scholarly source on Jiro Kawakita's KJ method, the originator of the technique. Article online.
- Gamestorming, Affinity Map: the run of the game, silent generation, participant-led sort and late naming.
- ASQ, Affinity Diagram (K-J Method): the quality-body reference, which places the technique among the seven quality management tools.