Why Workshop Clustering Fails (And How to Fix It)
The Problem No One Talks About
Clustering ideas after a divergent phase is something I've researched a lot over the years. Why? Because I think it's one of the hardest phases of any creativity workshop to get right.
Not ideation. Not convergence. Clustering.
That moment when you have 80 sticky notes on the wall and someone says "ok, let's group these" is where things quietly fall apart.
Why Clustering Fails
Here's the thing: clustering rarely fails because people are bad at it (even if untrained participants are indeed bad at it 🤣).
It fails because the facilitator does not set any clustering rules or strategy before the start.
What usually happens is the facilitator says "ok, let's group the ideas" and that's it. No lens. No focus. No shared logic.
So everyone does their own thing:
- Some people match by similar words
- Others group by general themes
- Others go by gut feel or visual proximity
You end up with tidy-looking groups that don't help anyone make a decision.
What Actually Works
What actually works is picking a clustering lens upfront and being explicit about it.
A clustering lens is a specific principle or perspective that guides how ideas are grouped together. Instead of clustering randomly or by surface-level similarity, a lens gives everyone a shared logic to follow: like grouping by root cause, stakeholder impact, or level of risk. It's the "rule" that makes clustering purposeful rather than arbitrary.
Ask yourself: what principle should guide how we group these ideas?
Here are some examples:
- By root cause. What underlying issue does this idea address?
- By who's affected. Which stakeholder or user group does this touch?
- By level of risk. How uncertain or risky is this idea?
- By trigger. What event or situation would make this idea relevant?
- By complementarity. Which ideas reinforce or complete each other?
It can be tricky to find the right grouping strategy. Ideally, you should prepare that before the workshop. But once you have it, say it out loud, write it on a flipchart, and keep people anchored to it throughout the clustering phase.
Why AI Won't Save You Here
A lot of facilitators are excited about AI clustering tools. And yes, whether you're using Miro AI or Stormz AI, they can group ideas fast.
But here's the catch: AI clusters by semantic similarity.
Semantic similarity is how AI measures if ideas "mean the same thing." Using embeddings (mathematical representations of text), AI calculates how close ideas are in meaning: like grouping "reduce costs" with "cut expenses" because the words are conceptually similar. It's fast and accurate for matching surface-level meaning, but it can't see deeper strategic connections or shared root causes that aren't obvious from the words themselves.
That sounds like what you want. But it actually misses the point entirely.
Sometimes the most valuable cluster is one that groups ideas that sound completely different, or even contradictory, because they share the same root cause or serve the same strategic goal.
Semantic similarity can't see that.
And I'm not even talking about the importance of having participants do the clustering themselves, for acceptance and consensus building. AI shortcuts that process entirely.
The Real Purpose of Clustering
The whole point of clustering is to help a group think better or move faster toward convergence and solution building.
If your clusters aren't doing that, they're just decoration.
Good clusters unlock decisions. Bad clusters just look organized.
What's Next
Right now, Stormz V2 uses embeddings for AI clustering—like most tools on the market. But the upcoming Stormz V3 will introduce a brand new approach that actually respects the clustering strategy defined by the facilitator, rather than just grouping by semantic similarity. More on that soon.