The Mental Models Opener
This room does not need introductions — everyone already knows where everyone has been. So skip them. Instead, each person puts one mental model on the table: a principle they actually use when they decide. Half an hour later you have something no icebreaker ever produced — a map of how this board thinks.
Set the rule before anyone speaks
Project this. Read it out loud, word for word, before anyone can slide into a biography.
The rule for the next half hour
We're not introducing ourselves. We're not talking about where we've been. Each of us shares one mental model — a central piece of how we think.
A mental model is a principle you hold true and actually use: a rule of thumb that decides things for you when the numbers are ambiguous. Name it as a short phrase, then say a couple of sentences about where you picked it up.
Today is a day about models — the ones in the machines and the ones in your heads. This is the second kind. Nobody needs to be profound; "never bet the company" counts.
Silence, on purpose
Start the clock and let the room sit with it. Resist the urge to fill the silence — the first minute is people discarding the polite answer.
While the clock runs
Name one principle you hold true — a rule you actually decide with.
Where did you pick it up? A career, a mentor, a mistake, a book. Two sentences is plenty.
One model per person, on the clock
Go around the table. Start the speaker clock for each person and capture their model as a card while they talk — short phrase, one line on where it came from. The clock keeps the storytellers honest.
Speaker clock
Capture the card
The model phrase is the only required part.
How this board thinks
Read it back to the room slowly. This wall is the day's reference map: when an AI question gets hard later, point at it and ask which of these models the room wants to apply.
Nothing pinned yet. Go back to the share round and capture each person's model as they speak.
How to run this — facilitator notes
Why this replaces the icebreaker
Board members who have sat together for years do not need to hear each other's CVs again. What you need — and what they have never heard — is how each person actually reasons. The exercise also primes the day's central theme: models as tools for understanding things you are not an expert in. They start the day by naming one of their own.
Timing plan for a board of 8–12
| 0:00–2:00 | Frame. Read the projected rule out loud. Take no questions — the exercise answers them. |
| 2:00–5:00 | Quiet thinking, clock projected. Do not talk. Do not check your phone; they will copy you. |
| 5:00–24:00 | Share round, ~2 minutes each. At 12 people, drop the speaker clock to 90 seconds and say so up front. |
| 24:00–27:00 | Read the wall back. One sentence of pattern-spotting, no more — the room should do the noticing. |
What to say at the key moments
"We're not introducing ourselves, and we're not talking about where we've been. Everyone here already knows that. Instead: one mental model — a principle you hold true, a cornerstone of how you decide. Name it, then two sentences on where you picked it up."
"Three minutes of silence first. The first thing that comes to mind is usually the polite answer — keep going past it."
After the wall: "This is how this board thinks. Every model up here is a tool for understanding something without being an expert in it. Keep that thought — it's the whole day."
What to listen for
Each card predicts how that person will react to the AI arguments coming later in the day. Rough translation table:
- Downside and accountability models ("skin in the game", "never bet the company") — this person will ask who owns it when the machine is wrong. Have your human-in-the-loop answer ready before they ask.
- Skepticism-of-abstraction models ("the map is not the territory", "trust the floor, not the report") — they will probe what the data misses. Give them the failure cases first; they will trust you more for it.
- Relationship and reputation models ("trust arrives on foot and leaves on horseback") — frame every AI use case in policy-member value and retention, never in headcount. Efficiency language loses this person.
- Consequence-chain models ("and then what?", second-order thinking) — they want the downstream effects: pricing, staff, regulator. Volunteer them before the pitch, not after.
- Evidence models ("base rates beat good stories") — demos will not move them; what happened at comparable insurers will.
- Reversibility models ("two-way doors") — your fastest allies for pilots. Frame the first AI experiments as cheap to undo and they will approve them in a sentence.
If two or three people converge on the same model, say so out loud — "this board runs on downside protection" is a diagnosis worth the whole exercise.
Running it solo
Working through this page alone? Do the exercise honestly for yourself: three minutes of thinking, then pin your own model. It is the fastest way to feel why the framing line matters — your first instinct will be to write your job title.