A pricing decision lands on your desk. One voice in the room says "let's work out what this should actually cost from the ground up." Another says "just match the market, that's what everyone does." A third says "it's basically the same play we ran on the last launch." All three are reasoning. None of them is reasoning the same way, and the disagreement underneath the meeting is rarely about the price. It's about which mode of thinking should win.

The quick version

  • First-principles reasoning strips a problem back to what must be true and rebuilds from there. Slow, expensive, and powerful when the inherited answer is wrong.
  • Heuristics are rules of thumb, fast, frugal shortcuts that are right often enough to be worth their speed, and biased in known ways.
  • Analogical reasoning carries the structure of a solved problem onto a new one ("this is like that"). Brilliant for insight, dangerous when the surface match hides a deep mismatch.
  • They're not a ranking. The skill is matching the mode to the stakes, the time you have, and how novel the problem really is.

The idea in depth

These three aren't competing theories of one thing; they're three different cognitive machines, each studied in a different corner of philosophy and psychology. Treating them as interchangeable is where leaders get into trouble, using a fast heuristic on a once-in-a-decade bet, or grinding through first principles on a decision that didn't deserve the afternoon.

First principles: demolish, then rebuild

The phrase is old. Aristotle used arche, a first principle from which other things are derived, and in Metaphysics Book IV called the law of non-contradiction "the most certain of all principles." Two thousand years later Descartes opened his Meditations on First Philosophy (1641) by resolving "to demolish everything completely and start again right from the foundations," doubting every belief until he reached something indubitable to build on. That move, tear down to bedrock, reconstruct only from what you're sure of, is the founding gesture of foundationalism in epistemology, and it's the same gesture a modern operator makes when they refuse the inherited answer.

So the move is: when a number or constraint is being treated as a law of nature, ask what it's actually made of. Elon Musk's much-cited battery example is the clean illustration, told that battery packs cost about $600 per kilowatt-hour and wouldn't fall much, he priced the raw materials (cobalt, nickel, aluminium, carbon, a steel can) at London Metal Exchange rates and got roughly $80, exposing the other ~$520 as convention, margin and manufacturing structure rather than physics. The discipline isn't genius; it's separating what must be true from what everyone assumes. Shane Parrish's Farnam Street guide offers two cheap entry points, Socratic questioning and the "five whys."

The limitation, honestly: first-principles reasoning is slow and error-prone if you get the principles wrong. Rebuild from a false foundation and you've laundered a bad assumption into a confident conclusion. It's also wasteful on routine choices, most decisions don't need a ground-up derivation, and pretending they do is its own failure mode.

flowchart TB
  A(["Inherited answer
('it costs $600')"]) --> B("Question the assumption")
  B --> C("Break down to
what must be true")
  C --> D("Rebuild from
verified facts")
  D --> E(["New answer
($80 in materials)"])
First-principles reasoning tears a claim down to bedrock, then rebuilds. Leaders Loop

Heuristics: fast, frugal, and good enough

Heuristics are the opposite trade. They don't rebuild anything, they substitute a hard question with an easy one. The modern study of them splits into two camps that are worth holding apart. Amos Tversky and Daniel Kahneman's landmark paper "Judgment under Uncertainty: Heuristics and Biases" (Science, 1974) catalogued shortcuts like representativeness, availability and anchoring, and showed they "lead to systematic and predictable errors." Their emphasis was the cost. Gerd Gigerenzer and the ABC Research Group, in Simple Heuristics That Make Us Smart (1999), pushed back: simple rules can be ecologically rational, they exploit the structure of real environments and, when information is scarce or the world is uncertain rather than merely risky, often beat more complex models.

Both are right, which is the useful part. A heuristic is a bet that the environment is stable enough for the shortcut to pay. In practice: name the heuristic you're actually using ("we anchor every quote on last year's number"; "we hire whoever reminds us of our best person"), then ask whether the environment still rewards it. Gigerenzer's distinction between risk (odds known) and uncertainty (odds unknown) is the tell. In genuine uncertainty a fast rule of thumb is often the smart play, not the lazy one. Where they fail is predictable: heuristics are confidently wrong in exactly the situations they weren't tuned for, and you rarely notice the misfit until the result is in.

A heuristic is a bet that the world is stable enough for yesterday's shortcut to still pay.

Analogical reasoning: this is like that

Analogy is how we think most of the time, often without noticing. Cognitive scientist Dedre Gentner's structure-mapping theory (1983) is the workhorse account: a good analogy maps the relational structure of a familiar "base" onto an unfamiliar "target," not the surface features. Douglas Hofstadter and Emmanuel Sander go further in Surfaces and Essences (2013), arguing analogy is "the fuel and fire of thinking", the core of cognition itself. The power is that a single good mapping transfers a whole web of relationships at once: understand how a thermostat regulates, and you've got a handle on a feedback loop anywhere.

So the move is: when you reach for "this is like X," interrogate the mapping. Are you matching deep structure, the incentives, the constraints, what breaks the thing, or just surface: same industry, same word, same vibe? The classic trap is the seductive surface match. "Uber for X" is the worn example, where the relational structure that made Uber work doesn't actually carry across. Gentner's own framing draws the line for you: ask what relations hold in the base, and check each one holds in the target. The weakness is built into the strength. Analogy persuades powerfully whether or not the deep structure transfers, which is exactly why it's the favourite tool of both great insight and bad strategy. (For the underlying logic of "argue from a parallel case," see our Toolkit piece on deductive, inductive & abductive reasoning, analogical inference is a species of inductive move.)

flowchart LR
  subgraph Base ["Known case (base)"]
    B1("Elements") --> B2("Relations
between them")
  end
  subgraph Target ["New case (target)"]
    T1("Elements") --> T2("Relations
between them")
  end
  B2 -. "map the structure,
not the surface" .-> T2
Structure-mapping: a good analogy transfers the relations between things, not the things themselves. Leaders Loop

A worked example

A SaaS company has to set the price of a new tier. Watch the three engines run on the same problem.

The heuristic in the room is "anchor on the competitor and shave 10%." It's fast and frugal, and for a commodity feature it might be exactly right, the market has already done the pricing work. But this tier isn't a commodity; it bundles something rivals don't have. The shortcut is tuned for an environment that no longer holds.

The analogy is "let's do what we did with the Pro tier, that launch went well." Useful, but the head of product checks the mapping rather than the vibe. The Pro launch worked because it converted existing power users with a clear before/after. The new tier targets a buyer the company has never sold to. Same company, same word ("tier"), different relational structure, the surface matches, the essence doesn't. The analogy survives as a source of questions, not as the answer.

First principles is the afternoon nobody wanted to spend. The team builds the price up: what value does this buyer actually capture, what does it cost to serve them, what's the lowest defensible floor, what would have to be true for the competitor's number to apply here? Illustratively, these figures are invented to show the mechanic, not real data, they find the feature saves a target customer roughly $2,000 a month, costs about $180 a month to serve, and that the "competitor minus 10%" anchor would have left most of that value on the table. The ground-up build doesn't hand over a final price either; it hands over the range the other two modes were silently narrowing without justification.

The payoff isn't that first principles won. It's that naming all three let the team see they'd been about to ship a heuristic answer to a first-principles problem, and that the analogy was smuggling in an unchecked assumption. The decision used all three: structure from first principles, speed from the heuristic where things genuinely were commodity, and the analogy as a checklist of what could go wrong.

flowchart TD
  A(["A decision arrives"]) --> Q{"Novel & high-stakes,
or routine?"}
  Q -- "Routine, stable world" --> H(["Heuristic:
fast rule of thumb"])
  Q -- "Resembles a solved case" --> N(["Analogy:
check deep structure"])
  Q -- "Novel, inherited answer
looks suspect" --> F(["First principles:
rebuild from scratch"])
  H --> R(["Decide, and note
which engine you ran"])
  N --> R
  F --> R
A rough triage: which engine the moment is asking for. Leaders Loop

Frequently asked questions

Is first-principles thinking just always the best one?

No. It's the most expensive, and it earns its cost only when the inherited answer is likely wrong and the stakes justify the rebuild. Run it on every decision and you'll be slow, exhausted, and no better than a good heuristic on the routine 90%. Reserve it for the bets where being conventionally wrong is costly.

Aren't heuristics just bias and laziness?

That's the Tversky–Kahneman half of the story, the part where shortcuts produce predictable errors. Gigerenzer's research is the other half: in an uncertain world, simple rules often outperform complex ones because they're robust to the noise that overfits a fancy model. A heuristic isn't lazy; it's a bet on a stable environment. The error is using one where the environment has changed.

How do I stop a bad analogy from steering a decision?

Make the mapping explicit. Write down the relations that made the base case work, then check each one holds in the target. Following Gentner, if you're matching surface features (same industry, same label) rather than structure (same incentives, same constraints, same failure modes), you've got a decoration, not an argument. "Uber for X" fails this test more often than it passes it.

Can I combine them on one decision?

You almost always should. Triage with a heuristic, sanity-check with an analogy, and reserve a first-principles rebuild for the part that's genuinely novel or where the conventional answer smells wrong. The worked example above does exactly this.

How is this different from deductive or inductive logic?

Those name the form of an inference (does the conclusion follow with certainty, or only with probability). These three name your strategy for generating an answer in the first place. They overlap, analogical reasoning is broadly inductive, but you can pick a strategy before you've fixed the logical form. See deductive, inductive & abductive reasoning for that layer.

Related in the Toolkit

Where to go next