You have met the manager who runs every problem through the same framework. Whatever lands on their desk, a stalled product, a tense hire, a budget shortfall, comes out the other side as the one thing they already knew how to solve. Sometimes they are right. Often they are confidently solving the wrong problem.

The quick version

  • A mental model is a simplified, reusable theory of how something works, supply and demand, compounding, incentives, feedback loops.
  • Charlie Munger's latticework idea: real understanding comes from holding many models, drawn from many fields, and hanging your experience on them.
  • Relying on one favourite model is the cognitive version of owning only a hammer, every problem starts to look like a nail.
  • The move is to build a small, deliberate kit of models from outside your discipline, and force every big decision through two or three lenses before you commit.

The idea in depth

The phrase most people associate with this topic comes from a 1994 talk Charlie Munger, Warren Buffett's business partner at Berkshire Hathaway, gave at the University of Southern California's business school, later anthologised in Poor Charlie's Almanack (edited by Peter D. Kaufman; reissued by Stripe Press in 2023). Munger's claim is blunt: "you can't really know anything if you just remember isolated facts." Instead, he argues, "you've got to have models in your head. And you've got to array your experience … on this latticework of models." His estimate, and he is candid that it is a rough one, is that "80 or 90 important models will carry about 90% of the freight in making you a worldly-wise person."

The latticework metaphor is doing real work. A list of models is inert. A lattice is a structure: the rungs cross-brace each other, so a new fact you encounter has somewhere to attach and several models to be checked against. The skill, then, isn't memorising more frameworks. It's catching yourself when you reach for a model and asking whether it's genuinely the right one or just the nearest one.

Why one model is dangerous: the law of the instrument

The failure mode has a name. In The Psychology of Science (1966), the psychologist Abraham Maslow wrote, "it is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail." (The underlying idea, the "law of the instrument", was stated a little earlier, in 1964, by the philosopher Abraham Kaplan.) Whatever discipline trained you, finance, engineering, law, sales, handed you a default model, and that model quietly pre-decides what you notice. An engineer sees a process problem; a salesperson sees a relationship problem; a lawyer sees a risk. They can all be looking at the same stalled project.

Worth treating your home discipline as a bias, then, and not only a strength. Before you act on the first framing that occurs to you, ask: what would someone from a completely different field see here? That single question is most of what cross-disciplinary thinking buys you.

flowchart TD
  P("A real problem") --> H("Your one default model")
  H --> S("A familiar-shaped solution")
  P --> L1("Economics: incentives?")
  P --> L2("Psychology: bias / motivation?")
  P --> L3("Systems: feedback loops?")
  L1 --> D("A decision you can defend")
  L2 --> D
  L3 --> D
One model funnels every problem toward a familiar answer; a lattice forces it through several lenses first. Leaders Loop

The evidence: foxes beat hedgehogs

This is not only folk wisdom from an investor. The political scientist Philip Tetlock spent roughly two decades tracking expert forecasts, in Expert Political Judgment (Princeton University Press, 2005) he reports gathering some 28,000 predictions from 284 experts and scoring them against what actually happened. His central finding: raw expertise barely predicted accuracy. What predicted it was thinking style. Borrowing Isaiah Berlin's image, Tetlock sorted experts into "hedgehogs," who know one big thing and extend a single grand theory into every domain, and "foxes," who know many small things and stitch together evidence from multiple traditions. The foxes were the better forecasters, and the gap widened on the harder, longer-range questions. Tetlock's later work on the Good Judgment Project (in Superforecasting, with Dan Gardner, 2015) found the same fox-like habits among the strongest amateur forecasters.

That gives the latticework an empirical spine: across thousands of real predictions, the multi-model thinker out-forecast the single-model expert. When a decision turns on a forecast, will this market hold, will this hire work out, will this launch land, distrust the confident one-theory answer and reward the one that arrives wrapped in "howevers." This connects directly to how you choose a reasoning approach in the first place: latticework thinking is, in part, the discipline of not defaulting to a single one.

"80 or 90 important models will carry about 90% of the freight.", Charlie Munger, USC, 1994

An honest limitation

Two cautions keep this from becoming its own cliché. First, Munger's "80 to 90 models" is an estimate offered in a speech, not a measured result, treat it as a vivid target, not a law. Second, more models is not automatically better. A shallow tourist's grasp of twelve disciplines can produce confident nonsense, fluent analogies that don't hold. Tetlock's foxes weren't dilettantes; they were disciplined about checking their many models against reality and updating. The point isn't to collect frameworks like badges. It's to hold a few you actually understand, from a few different fields, and to know when each one applies and when it doesn't.

A worked example

A regional sales team is missing target for the third quarter running. (The numbers here are illustrative.) The VP of Sales, a hedgehog by training, reaches for the only model that has ever served them: activity drives results. The plan writes itself: more calls, a steeper commission curve, a leaderboard on the wall. It is the hammer, and the quarter looks like a nail.

Now run it through a small lattice instead. Through an economics / incentives lens (a model you can borrow from basic economic reasoning): the steeper commission may simply pull reps toward easy renewals and away from the slow new-logo deals that the target actually depends on, the incentive is fighting the goal. Through a psychology lens: a public leaderboard motivates the top two reps and quietly demoralises the bottom six, who stop trying. Through a systems-thinking lens: there's a feedback loop, fewer wins this quarter means thinner pipeline next quarter, so "work harder now" borrows from a future that's already short.

Three lenses, three different root causes, none of them visible from inside the activity model alone. The defensible decision probably isn't "more calls." It's: fix the comp plan so it points at the deals that matter, replace the leaderboard with coaching for the middle of the team, and protect pipeline-building time so the loop stops draining. Same data, better decision, because it was forced through more than one model before anyone committed.

flowchart LR
  M("Missed target") --> A("Activity model: do more")
  A --> R1("Risk: wrong activity, faster")
  M --> B(["Run the lattice"])
  B --> C1("Incentives: comp vs goal")
  B --> C2("Psychology: leaderboard effect")
  B --> C3("Systems: pipeline feedback loop")
  C1 --> F("Fix comp · coach the middle · protect pipeline")
  C2 --> F
  C3 --> F
The same shortfall, read through one model versus three. Figures and scenario are illustrative. Leaders Loop

Frequently asked questions

Isn't this just "think outside the box" with extra steps?

No, it's more specific and more testable. "Think outside the box" tells you to be creative without telling you how. Latticework gives you the mechanism: keep a named set of models from different fields, and deliberately apply more than one to each big decision. Tetlock's forecasting data shows the multi-model habit measurably outperforms the single-model one. That's a method, not a mood.

How many models do I actually need?

Far fewer than Munger's eighty to start. A working leader gets most of the benefit from a handful that recur everywhere: incentives, supply and demand, compounding, feedback loops, opportunity cost, base rates, and a few cognitive biases. Master five or six you genuinely understand before you go collecting more, depth in a few beats a shallow tour of fifty.

Where do I find good models without doing a second degree?

Borrow the big load-bearing idea from each field rather than the whole field. Economics gives you incentives and opportunity cost; biology gives you evolution and ecosystems; engineering gives you margin of safety and feedback; psychology gives you the common biases. One strong idea per discipline, properly understood, is the 90% Munger was pointing at.

Doesn't running everything through three models just slow me down?

Reserve it for decisions that are expensive to reverse. Most choices are cheap and reversible, make those fast on a single sensible model. Latticework thinking earns its time on the few big, hard-to-undo calls, which is exactly where one-model tunnel vision costs the most.

How is this different from first-principles thinking?

They're complementary. First-principles reasoning rebuilds a problem from the ground up; latticework checks it from many angles. You can use first principles to understand one model deeply and the lattice to make sure you're not trapped inside it. See first principles vs heuristics vs analogical reasoning for the distinction.

Related in the Toolkit

Where to go next