Affinity mapping: cluster ideas into themes that decide

Affinity mapping is the method of sorting a large set of ideas or observations into clusters of related items, then naming each cluster as a theme. It turns 40 to 100 scattered notes into 5 to 8 named groups you can count, compare and decide between.

The method comes from Jiro Kawakita's KJ technique, built for field researchers drowning in unsorted observations, and it earned its place in every design and product toolkit for one reason: it is the bridge between generating and deciding. A brainstorm gives you a wall of sticky notes. Affinity mapping tells you what the wall says.

How to run affinity mapping, step by step

1. Get everything into atoms. One idea per note, written so it stands alone. "Improve onboarding" is not an atom; "send a setup checklist email on day 2" is. Compound notes must be split before sorting or they will glue two themes together.
2. Spread everything out unsorted. Physical wall or digital canvas, all notes visible at once. Resist pre-sorting as you place them; the value comes from seeing the whole pile before any structure exists.
3. Sort in silence first. Each person moves notes next to notes that feel related. No discussion for the first 10 minutes: talking this early lets the most confident voice define the categories before the pattern has a chance to emerge on its own.
4. Let clusters form bottom-up. Do not create category labels and file notes into them. That is sorting into buckets you already believed in. Real affinity mapping lets the groups emerge from the notes, including groups nobody expected.
5. Name each cluster with a sentence, not a noun. "Pricing" is a filing label. "Customers cancel because they cannot predict their bill" is a theme, and it already suggests what to do. If you cannot write the sentence, the cluster is not a real theme yet.
6. Split anything over roughly 10 notes and keep the orphans. A giant cluster is usually two themes wearing one label. And a note that fits nowhere is not trash; outliers are often the only genuinely new thought in the room, so park them in a visible "unclustered" group.
7. Count and record. Write the tally on each theme. The map decays fast; photograph or export it, with counts and theme sentences, the same day.

Budget 20 to 30 minutes for 50 to 80 notes with a group of 3 to 6 people. Solo, it works fine and often faster; you lose the silent-sort ritual but keep the core mechanic. Generating the notes is the part that compresses: Brainstormer returns around twenty ideas already tagged by the persona that produced them, which gives you a first cut to sort rather than a blank wall. The grouping itself stays yours, and our idea clustering guide covers how to do it well.

How do you name affinity map themes?

Name each cluster with a short phrase that states what the notes in it have in common, written as a claim rather than a category. "Users cannot tell which plan they are on" is a theme name. "Billing" is a filing label. The test is whether someone who never saw the raw notes could act on the name alone, and category words almost never pass it.

Four rules make the difference between names that decide things and names that just tidy the wall:

• Write a sentence fragment, not a noun. Nouns like "Onboarding", "Pricing" or "Performance" describe where a theme lives, not what it says. They feel neutral and safe, which is exactly the problem: a neutral name can be agreed with by everyone and acted on by nobody.
• Name it after the reading, not the words. The name should capture what the group concluded the notes mean, which is usually more specific than any single note. If three notes say "confusing", the theme is not "confusion", it is whatever the three of them were confused about.
• Name it last, and out loud. Cluster first in silence, then name each group as a short discussion. Naming while sorting locks in the first label somebody says and quietly bends later notes toward it, which is how you end up with a "Miscellaneous" pile of things that did not fit a premature name.
• If you cannot name it in one line, it is two themes. A name that needs an "and" in the middle is the clearest signal a cluster has merged two distinct ideas. Split it and name the halves separately. The reverse is also true: two clusters you keep wanting to give nearly the same name are one cluster.

Watch for the "Other" pile. It is legitimate during sorting and a problem at the end, because unnamed notes get dropped from every decision that follows. Before you finish, either place each stray note in a named theme or accept that it is a genuine one-off worth keeping visible on its own, which is sometimes exactly what the most interesting note is.

Reading the counts: what cluster sizes actually tell you

The counts are evidence, but of a specific thing: they measure where the group's attention went, not where the value is.

• A big cluster (10+ notes) means the theme was easy to think of. Sometimes that is because it matters; often it is because it is familiar. When a SaaS team affinity-mapped 60 churn ideas, 22 landed in "improve onboarding emails," the thing they already discussed weekly. Big clusters deserve attention and suspicion in equal measure.
• A mid-size cluster (4 to 8 notes) reached by several people independently is the strongest signal in the map. Independent convergence is worth more than raw volume, especially if the notes arrived from different roles: when support, sales and engineering each wrote a note about billing surprises, that theme won despite ranking third by count.
• A small cluster (2 to 3 notes) is an underexplored direction, not a loser. The correct response is often another divergence round aimed only at that theme.
• Orphans are single notes nobody could place. Review them last, explicitly. Most are noise; occasionally one is the reframe that makes the whole map look different.

Cluster size measures how easy a theme was to think of, which is not the same as how much it matters.

Common affinity mapping mistakes

• Pre-built categories. Filing notes into "Product / Marketing / Support" buckets you named in advance guarantees you learn nothing; you already knew those words. Let the clusters emerge, then name them.
• Sorting by topic instead of by meaning. "Both notes mention email" is topic. "Both notes are about customers not knowing what to do next" is meaning. Topic-sorted maps look tidy and decide nothing.
• Noun labels. A theme named "Onboarding" hides three different problems. Force the full sentence.
• Talking during the first sort. Early discussion converges the room prematurely, the exact failure that separating divergent and convergent thinking is meant to prevent.
• Forcing every note into a cluster. A map with zero orphans is a map someone tidied too hard.
• Treating counts as votes. Fifteen variations of one obvious idea should not outrank four notes describing a real, unserved problem. Counts start the conversation; they do not end it.
• Stopping at the map. The most common failure: the wall gets photographed, admired and abandoned. A map that does not feed a decision was a team-building exercise.

From affinity map to decision

The map's job is to shrink the decision space from 60 ideas to a handful of themes. Finish the job in three moves. First, pick 2 or 3 themes to act on, using the counts plus the independence signal, and say out loud why the big familiar cluster did or did not make the cut. Second, within each chosen theme, score the individual ideas on impact against effort; the theme told you where to dig, the scoring tells you which shovel. An impact effort matrix is the fastest version of that second move, and our idea prioritization guide covers how to run it across the whole wall. Third, write the decision in one line ("we test pause-instead-of-cancel this sprint because the billing-surprise theme had cross-team convergence") and archive the map with it, so the reasoning survives the week.

This diverge, cluster, score, commit arc is the whole ideation process in miniature, and affinity mapping is its hinge. Run the sort honestly, read the counts skeptically, and the wall of notes becomes what it was always supposed to be: one defensible next step.