Affinity Diagram

Often you'll have too many ideas. You may use cardstorming to elicit ideas from workshop participants and come up with hundreds of ideas. You might interview users or visit the site where software is being used and take copious notes. If notes are arranged one idea per index card, it's easy to end up with hundreds of these cards. Distilling a large number of ideas into themes that help you understand the information can be extremely difficult.

A single individual may look at a large amount of information, see patterns and form ideas about what's important. But, any subsequent person who looks at the same information may see different patterns. It's important to model the information in a way that everyone can see the same themes and make similar decisions about what's important.

An affinity diagram lies at the foundation of many modeling techniques. It's a simple idea that the human head seems particularly good at. Simply arrange ideas that seem similar to each other, ideas that have affinity, closer together. Arrange ideas that or dissimilar farther apart.

Affinity diagrams are very helpful to distill large amounts of information and find higher order concepts and themes in that information.

To build an affinity diagram:

1. Start with ideas written on index cards or post-it notes, one per card or note. (We'll need to arrange the ideas individually, so don't try to save paper here.)
2. Place the ideas down on a tabletop one at a time.
3. Place ideas that seem similar to each other close together, ideas that seem dissimilar farther apart.
4. As the model clumps you may need to move groups of ideas around the models. This is where index cards that slide easily become valuable.
5. When the model settles you should have clusters of cards that are similar. For each cluster, consider why it is similar. Place a different color index card or post-it on the cluster. Write on that card a few words that summarize or distill the information found in the cluster of cards under it.

Ta daaa! You've built a simple affinity diagram.

Try this using your next list of anything: your to-do list, your shopping list, the activities on your current project, or your goals in life. I promise you'll learn something about the information by doing this. The larger the number of ideas, the more you'll learn.

Building an Affinity Diagram Collaboratively

Building affinity diagrams in a collaborative setting generally works well. Everyone should listen closely to the conversation that occurs while participants are clustering cards. This is where you'll find disagreements or misunderstandings about the words written on cards.

Listen closely while participants cluster cards to understand why they're clustering. Try to understand exactly what affinity means. This discussion continues to help build common understanding.

If there are large numbers of cards to model, divide the cards up among collaborators giving each a stack to place. You'll find you can get through a large number of cards fairly quickly.

With large models, you'll start to see themes emerge well before all cards are placed. It's okay to begin to label clusters with different colored cards before all cards are placed. This may help speed up placement.

Do split up large clusters. Some things may seem to have affinity early on, but as more cards are added to a cluster, its "personality" may change. Pull cards out that no longer seem to fit and let them start their own cluster close by.

Ad Hoc Isn't Random

While placing cards into clusters, something interesting and unspoken seems to be at work. People naturally place important ideas higher up or closer to the top edge of the workspace than less important ideas. If the ideas concern activities that happen over time, many people tend to locate ideas that happen early to the left of ideas that happen late. If many ideas rely on one idea or group of ideas, often a radial arrangement results where a single idea or cluster is in the center of the model, and all other ideas or clusters are positioned around that idea.

Allow clustering to happen naturally, but don't consider the clustering random. When the model is complete, consider why ideas and clusters appear where they do. There often are specific reasons.

Clustering: Blobs, Skyscrapers, and Cityscapes

The information in individual clusters can be seen as small models in their own right.

I've observed some groups arrange clusters into arrangements that seem to have no target shape - more like a blob where cards overlap each other to form strange shapes. Often these blobs have their own organization where important ideas or clarifying ideas rise to the top of the blob. Or, alternatively the blob may be radial - where the center of the blob contains the critical ideas.

Alternatively I've observed groups arrange ideas into neat vertical columns or less common horizontal rows of cards where each column or row is a cluster. While I naturally tend to create blobs, I've found that in practice these columns have some advantages.

Columns can be a little more compact - by overlapping cards you can position them where all text is visible, but they occupy less space.

Since the cards are shaped like a list, it's easy to treat them as one, and arrange the ideas in a rough order of priority or importance.

It's easy to cap a column of cards with a different color card to summarize it.

Since the arrangement is regular and predictable, it's easy to gather cards up into a stack and rubber-band them. The regular arrangement of cards can be accurately reassembled later on.

When lots of columns of cards are placed together left to right, they tend to look like skyscrapers. The entire model tends to look like a cityscape.

Splitting and Combining

As you review clusters and the ideas inside them, you may find it useful to split a cluster into two, or combine two closely related clusters into one.

Hierarchies

The simplest of affinity diagrams is a two level hierarchy: the clusters the bottom level, the labels on the clusters the top level. You needn't stop with two levels.

You may find that several clusters have affinity with each other. Arrange these clusters close together then label the cluster of clusters. By doing so you've gone up a level of abstraction and increased the depth of the hierarchy.

You may find that large clusters - more than five or six cards - have themes inside them. Break the cluster into one or more smaller clusters. Label each of these smaller clusters. By doing so you've added a "mezzanine" level of abstraction - one above the lowest level of detail contained in the cards but below the level of detail originally summarized in the label for the cluster before splitting it.

Fixing & Annotation

As you discuss an affinity diagram, you'll rearrange clusters, split, merge, form hierarchies, and discuss details about each cluster and their relationships with each other. At some point the model will begin to settle. By now you'll realize that both the information contained in the cards AND their spatial arrangement is important.

I'll often start a collaborative worksession by covering a tabletop with poster paper. At the end of an hour or two of modeling I need only pull out the tape and fix the cards to the paper already on the tabletop. Then I can carefully roll up or fold up the model and carry it away.

While there's lots of unspoken information in the card arrangement, sometimes it's important to make some ideas explicit.

Make notes directly on the model about conversations that occurred during the session. Draw lines connecting clusters of information with each other and label the lines with text describing the nature of their relationship. Title the model. Write the names of the workshop participants and the date the model was constructed. Draw diagrams or silly pictures that help remind everyone of the conversations that occurred while the model was being created.