Affinity diagram, clustered.
Also called: KJ method · Affinity mapping · Theme clustering · Bottom-up grouping
Sorting scattered research notes into natural themes, so the patterns surface from the notes themselves rather than being decided in advance.
Write every observation on its own card, then group the cards by what feels related until themes emerge on their own. Think of sorting odd socks into piles by feel, not by a list you wrote first. The patterns lead; you follow. There is a worked cluster set below.
What an affinity diagram is
An affinity diagram is a way of making sense of a pile of messy research. You take everything you heard in your interviews and observations, write each separate point on its own card or sticky note, and then move the cards around a table until ones that feel related end up together. You do not name the groups first. You let the groups form, and only then ask what each one is really about.
The reason the order matters is simple. If you decide your themes up front (price, design, features) and then sort notes into those bins, you only ever confirm what you already believed. Sorting bottom-up is the opposite move. A theme you never expected can announce itself because six unrelated-looking notes keep drifting into the same corner of the table. That is the whole point of the method: it shows you what is actually in the room, not what you went looking for.
Here is the analogy I keep coming back to. It is like sorting a basket of odd socks. You do not start with labelled boxes; you start pairing by feel, by colour, by thickness, and the piles tell you what kinds of socks you own. The same thing happens with research notes. The piles teach you the categories, instead of the categories forcing the piles.
When we worked through the interview notes from Discover on the proofing box, we ended up with a small set of clusters that no one would have written down as headings on day one. Here is the shape of what emerged, so you can see a real result rather than a generic template.
Notice that the last cluster, the distrust of apps, was never a question we set out to ask. It assembled itself from offhand remarks across four separate interviews. A pre-decided set of buckets would have buried those remarks under “features” and we would have missed the strongest signal in the whole study.
The difference between sorting bottom-up and sorting into bins you wrote first is the difference between learning something and confirming your own assumptions.
- You write the headings first: price, design, features.
- Every note gets filed into a bin you already believed in.
- Surprises get flattened into the nearest tidy label.
- You finish the day confirming what you walked in thinking.
- You write down what people actually said, one point per card.
- Cards drift together by feel; no labels yet.
- A theme nobody expected can declare itself.
- You name each group only after it has formed.
How it fits the bigger picture
Affinity diagram is activity 03.02 in the framework, early in Stage 03 Innovate. It takes the raw notes gathered in Stage 02 Discover and turns them into themes you can act on, which then feeds straight into define audience (03.03), where those clusters become a clear picture of who you are really building for.
What it can do
It surfaces patterns you would otherwise miss, because it lets the structure rise out of the notes instead of being imposed on them. Done honestly, it turns a wall of disconnected quotes into three or four themes the whole team can point at and agree on.
What it can’t do
It can’t tell you how big each theme is or which one to prioritise; that is what define audience and the later Stage 04 Evaluate activities do. The clusters are a sorted picture of what you heard, not a ranked plan, and a cluster of one note is a hunch, not a finding.
See the full 10-stage process →
Try it yourself
Take your raw notes from Discover. Write each separate observation on its own card, in the person’s own words where you can. Spread them out and start pairing the ones that feel related, with no headings yet. When a pile has formed and stops growing, only then ask: “what is this group actually about?” and write that name on top. Aim for three to five named clusters; if you have a dozen, you are grouping too finely.
Or run the guided version. The Free Sprint walks you through turning research into themes as part of its discovery questions. Start the Free Sprint →
Your affinity-diagram checklist
Project notes: the cluster nobody planned
▸ From the notebook · optional reading
Clustering the Discover interview notes with Dan and Anna Hartley in Stockport, and the theme that quietly decided the no-app design.
3 min read · click to open
We came out of Discover with about forty pages of interview notes and a strong urge to skip straight to “so it needs an app”. I asked Dan and Anna to hold off. We spent an afternoon at their kitchen table in Stockport with a stack of index cards, writing one observation per card, in the baker’s own words where we had them.
How the piles formed
We did not label anything. We just started moving cards into loose piles by feel. Within an hour the table had sorted itself into four rough groups. Three were roughly what you would expect: a fear of wasting a bake, no room for another gadget, and wanting the thing to look good on the counter.
The fourth one I genuinely did not see coming. A small pile kept growing out of throwaway lines: “I’m not setting up another account for a bread tin”, “does it nag me?”, “I just want a dial”. On their own each was a shrug. Stacked together they were the loudest signal in the study: a quiet distrust of apps on kitchen kit.
What the cluster decided
That pile is the reason the box has no app, no Wi-Fi and no account. Had we sorted the notes into bins we wrote first, those lines would have been scattered under “features” and “nice-to-haves” and we would have missed it entirely. Instead it became a design principle. I remember Anna looking at the table and saying it was obvious now, which is exactly how a good affinity cluster feels: invisible until it isn’t.
We carried those four named clusters straight into define audience the following week. They did most of the work for us.
— Innovate stage, project notes, 2026
— Next in Innovate → Define audience