No caveats - so far over the years I have been using this, sceptics and even those who know the technique and have a feeling about what will be revealed, all use it willingly and correctly.
Cards, Post-Its, or similar small, low fidelity things that are easy to pick up, have some information on them, and place somewhere else (for brevity, I will refer to these as just cards)
Each card contains either some writing (eg data collected from team members during a retrospective, news articles) or some graphic (eg pictures, photos,
Archetypes, etc). Spread all the cards out and get the group to have a look/read through them in a
timebox.
Depending on the size of the group, the number of data cards, 5-10-15 minutes should be sufficient for each timebox. It is the pressure of the timebox that drives the right outcomes in a group setting, but for 1-1 sometimes a little less pressure is more beneficial - be guided by your coachee!
Step 2:
Again setting the timebox, ask everyone to cluster: everyone has to find cards, attempt to keep them on the surface or near, 1-at-a-time, and move that card closer to one or more others that the card is related to in some way either obvious or not. Essentially we're facilitating the creation of categories without names (so far) that are important to this group, in this data, in a consensus building way.
Step 3:
Often the wall or table is too small for the group and/or number of cards, and often some members of the group like to stand back and watch, rather than participate. Your job as facilitator is to make sure every person moves several of the cards, and participates/contributes to the creation and destruction of the clusters. I often just use a "line up" method and request people to move back as I think they've had enough time, and request people to move forward as I think they have not had enough time.
A major team-building side-effect/benefit of a small room/surface at this time, is that people actually have to talk to each other and negotiate with verbal or non-verbal cues as they try to move past each other - which all adds up to better team understanding and cohesion! Of course this adds to Health and Safety concerns, so be sensible.
Step 4:
Once the clusters are settled, as facilitator you need to quality check them. Are all the bugs in the bugs cluster? Are all the specification problems in the specification problems cluster? And all the risks are in the risk cluster? And all those others we're unsure what to do with, they're in the miscellaneous cluster? Right?
WRONG! However, to extricate from the above incorrectness, a useful way is to take each miscellaneous card and recluster as much as possible from the unuseful clusters around that.
Alternatively, request the group to try again, this time finding other relationships such as timing, specific people, specific technology failures, specific bug etc etc ... and see what cluster data then appear. And then ask them to try again, and again, until really, there are some real and new themes that have emerged!
Step 5:
Now ask the group to collaboratively name the clusters according to the following rules:
1. Short concise descriptive sentences are better than
2. Short descriptive phrases are better than
3. A list of words are better than
4. A single word
And rotate the writer of the cluster name for each cluster. The group has reach consensus - again a great team building activity to practice in low-risk non-crisis moments! The name of the cluster is written typically on a different colour card, or with a different colour pen, or font, and is usually underlined to distinguish it clearly from "data cards".
If a good name can not be agreed on by consensus, often it means the cluster should be broken down again, and this is a good thing! People, especially in fast changing environments, learn more from retrying than just getting on with old themes, old behaviours, old attitudes. This is a high energy, exciting, engaging, positive, great thing to do, not boring, unless you're letting the person or group members do it wrong. A little uncertainty, discomfort goes a long way to really getting to the heart of [hidden] matters.
At this stage, for the coaching technique of clustering, we're done. We have allowed underlying themes hidden in data cards, to emerge into categories that are somehow, and some why, important to the individual you're coaching or the team you're facilitating, who have been making consensus after consensus, and not really trying to out-think the future consequence of the enjoyment they're getting from this physical exercise. And if you then "litmus test" the discovered themes, generally the coachee or group members agree they would never have suggested those as areas where there is some weakness or strength to be explored further with goal setting and plan creation to achieve.
must have book, as well as Luke Hohmann's 
, and seen just what gold clustering produces!
which I also use and think is great for complex ad hoc charts!)