Group
Cluster related and duplicate nodes, and park what does not fit, without losing any contribution.
Purpose
Reduce the noise on the board before anyone draws an arrow. Stack duplicate nodes, move related nodes next to each other, and use the parking lot for nodes that do not fit. Nothing anyone wrote should be deleted.
Clusters make the next step easier. When related nodes sit together, the connections between them are easier to see.
When it starts and ends
Group starts when Seed closes. It ends when duplicates are stacked, related nodes sit together, and the parking lot holds whatever does not fit yet.
A small board groups in a few minutes. Even then, do it: every step happens in every session, however briefly.
Steps
-
Read the nodes out loud, or give everyone some time to scan the board in silence.
-
Ask for duplicates and relatives.
What duplicates do we have? What's related?
-
Stack duplicates, do not delete them. Place each duplicate behind the node it repeats, slightly offset, so both stay readable. If 2 nodes say nearly the same thing, the group may write 1 summary node and keep both originals stacked under it.
-
Move related nodes next to each other in loose groups. Label a cluster if the label helps people find their way. A label is a reading aid, not a category.
-
Solutions, out-of-scope items, and premature details go to the parking lot, and their authors hear why:
This is a good point. Let's park it for now and come back to it if we need it.
-
Look across the clusters. Before moving on, ask whether clusters the group treats as separate might be connected. This is the convergence pattern, and it often finds the most valuable arrows on the board.
-
Close Group, and move to Connect:
Now let's start connecting things. Who influences what?
Timing
| Group size | Suggested time | What to expect |
|---|---|---|
| 3-7 participants | 5-10 min | Small boards de-duplicate fast |
| 8-15 participants | 10-15 min | More duplicates, and some merge decisions |
| 16-30 participants | 15-20 min | Split the board among small groups |
For the whole session, see timing.
Tips
- Let participants move the nodes. Authors know which of their nodes belong together, and moving them keeps everyone involved.
- Split a large board. With 16 or more participants, give each small group one area of the board to cluster, then walk the result together.
- Keep clusters loose. Their only job is to make Connect easier. Once arrows carry the structure, the cluster frames come off.
- Treat the parking lot as a promise. Everything parked gets a second look at the end of the session, during Close.
Traps
- Deleting contributions. A node that disappears silently tells its author that their input did not count, and they stop contributing. Stack it or park it instead.
- Using the parking lot to make nodes disappear. A node parked without a word is a node deleted. Say why it is parked, every time.
- Debating the wording. Which duplicate is phrased better does not matter yet.
- Forcing categories. Not every node belongs to a cluster. A node may sit alone until Connect shows where it fits.
Example
The support corner of the renewals board continues from Seed.
"Everything ends up with engineering" went behind "Most tickets are escalated to engineering," the pre-seeded node it repeats. An engineer pointed out that tickets about custom setups always end up with engineering, so "Large customers run custom code paths" joined them in a cluster the group called "Escalations." The 2 nodes about what tier 1 can and cannot see formed "Support tooling." The facilitator moved "Buy a new helpdesk tool" to the parking lot, thanked its author, and said the group would ask what problem it solves once the arrows were in.
Step 1 of 3. The corner at the end of Seed.
Step 2 of 3. Related nodes move next to each other. The duplicate is stacked on the node it repeats.
Step 3 of 3. The solution goes to the parking lot, not the bin. Nothing anyone wrote is lost.
Diagram as text: The support corner of the renewals board during Group
Steps:
- Step 1: The corner at the end of Seed.
- Step 2: Related nodes move next to each other. The duplicate is stacked on the node it repeats. Changes: "Escalations" holds "Everything ends up with engineering", "Most tickets are escalated to engineering", "Large customers run custom code paths"; "Support tooling" holds "Tier 1 support cannot see a customer's setup", "Engineering's internal admin tool".
- Step 3: The solution goes to the parking lot, not the bin. Nothing anyone wrote is lost. Changes: "Parking lot" holds "Buy a new helpdesk tool".
After the last step:
Nodes:
- Root / Problem: "Most tickets are escalated to engineering" (in "Escalations")
- Desired State / Outcome: "Tier 1 resolves 7 in 10 tickets on its own"
- Desired State / Outcome: "Engineers spend 80% of their week on planned work"
- Not yet typed: "Everything ends up with engineering" (in "Escalations")
- Not yet typed: "Large customers run custom code paths" (in "Escalations")
- Not yet typed: "Tier 1 support cannot see a customer's setup" (in "Support tooling")
- Not yet typed: "Engineering's internal admin tool" (in "Support tooling")
- Not yet typed: "Buy a new helpdesk tool" (in "Parking lot")
No arrows yet.
Across the whole board, 5 clusters formed: account health, support and escalations, deals and roadmap, releases and reliability, and incident communication. The parking lot held 3 nodes, among them the helpdesk tool and a node about pricing, which is out of scope.