Shared Mental Models: Why Experienced Teams Need Fewer Words

TwoTwo developers who've shipped three projects together can fix a production incident over a two-word exchange — "cache again?", "on it" — that would mean nothing to a new hire listening in. The new hire isn't missing information about the bug. They're missing the thing that made two words sufficient in the first place: a model of the system, and of each other, close enough that most of what would normally need saying doesn't need saying at all.

What a Shared Mental Model Actually Is FoundationalKnowledge that endures for decades — core principles

JanisJanis Cannon-Bowers, Eduardo Salas, and Sharolyn Converse gave this phenomenon a name and a mechanism: a shared mental model is a knowledge structure held in common across team members, one that lets each person form accurate expectations about the task and about each other, and coordinate action on the strength of those expectations rather than through constant explicit negotiation¹. The claim isn't that experienced teammates read each other's minds. It's that they've each separately built a model of the task and of each other accurate enough that the same situation reliably produces compatible expectations in both heads at once, which is what makes a two-word exchange carry as much coordinating weight as a new pairing's two-minute one.

Novices Need Instructions; Experts Need a Nod Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

ThisThis is why onboarding a new team member always feels slower than the task itself should warrant, and why the slowdown isn't really about the task. A new hire doesn't yet have a model of how this particular team's incidents usually unfold, which teammate tends to check the cache first out of habit, or what "on it" is shorthand for in this specific group's history of fixing this specific class of problem. None of that is written down anywhere, because none of it needed to be — it accumulated the way a tailoring apprentice's feel for a seam accumulates, through repeated shared exposure to the same kind of situation, not through a briefing. An incident-response call with a well-gelled team sounds almost silent by comparison with the same call run by a team thrown together for the first time, and the silence is the shared model doing its job, not a sign that less is happening.

Can a Misunderstanding Be Collective Rather Than Individual? FoundationalKnowledge that endures for decades — core principles

TheThe same mechanism that lets two words carry a correct diagnosis can let two words carry an incorrect one, just as efficiently, and this is the sharper edge of the idea worth drawing out explicitly, since the framework as usually presented emphasises the coordination benefit without dwelling on this cost. If both halves of a pairing have independently converged on the same mistaken model of a system — a cause that was true two deployments ago and no longer is, say — their smooth, low-word coordination reproduces the error faster than it would expose it. A newcomer asking "wait, why do we think it's the cache?" is a nuisance to a team running on a shared model, and is also the only member of the exchange in a position to notice the model is wrong. A team's efficiency at using a shared model and its accuracy in having built the right one are two separate properties, and the first can look identical whether the second is true or false.

Strongest Objection: Shared Models Take Exactly the Time They Take FoundationalKnowledge that endures for decades — core principles

NothingNothing about this mechanism is fast to build. A shared mental model accurate enough to compress "why do we think it's the cache" down to "cache again?" is the product of the same slow, repeated, co-present exposure to shared situations that legitimate peripheral participation describes for expertise generally — which means a team assembled quickly, for a short-lived purpose, or working mostly asynchronously starts with none of it and has limited time to build any. What a temporary team can do instead, when it doesn't have the luxury of years together, is this cluster's own next question.

Provisional Conclusion FoundationalKnowledge that endures for decades — core principles

AA shared mental model is a real, specific thing a team can have more or less of, not a vague way of saying "they work well together" — it explains concretely why experienced coordination compresses toward silence rather than staying verbose as a team improves, and it explains, just as concretely, why that same compression offers no protection against a shared model that's simply wrong. Building one takes the kind of time a fast-forming team rarely has, which is exactly the gap the next page in this cluster opens with.

Questions for Further Thought

• Is there a reliable external signal that a team's silence reflects a well-converged shared model rather than a wrongly-converged one, before the model is actually tested by a real failure?
• Can a shared mental model be deliberately built faster than the slow, repeated-exposure route Communities of Practice describes, or does the mechanism require that much real shared history regardless of method?
• Does adding a newcomer to a team with a strong shared model do more good by eventually absorbing it, or by temporarily disrupting it — by being the one person still asking "why do we think that"?

Related Topics

• Can an Organisation Know Something No One in It Knows? — the distributed-processing claim a shared mental model is one concrete mechanism for.
• Communities of Practice: Learning by Belonging, Not by Being Told — the slow, repeated-exposure route by which a shared mental model actually gets built.
• Multi-Agent Systems: Coordination, Communication, and Strategic Interaction — the equivalent coordination problem for artificial agents, where an explicit shared model is usually engineered rather than slowly absorbed.
• How Much Memory Does a Temporary Team Need to Become a Community? — what a team without time to build a shared model relies on instead.

Further Reading

• Cannon-Bowers, J. A., Salas, E., & Converse, S. (1993). Shared mental models in expert team decision making. In N. J. Castellan Jr. (Ed.), Individual and Group Decision Making: Current Issues (pp. 221–246). Lawrence Erlbaum Associates.

References