First Principles Thinking

The problem it solves

Most of the constraints a team works under are somebody else's conclusions, reached under conditions that have since moved. The release window is a fortnight because it was a fortnight at the last company, the pricing has three tiers because three tiers is what the category does, and the extra approval step exists because something went wrong once for people who have all since left.

Copying an answer that works is usually the right move, and reasoning by analogy is fast, cheap and correct far more often than it's wrong. What it can't do is reach an option nobody in the category has taken, and it quietly imports the assumptions of whoever produced the original, none of which arrive labelled.

The tell is a decision that keeps coming back with the same short list of options, each one a variation on what's already being done, and an argument that runs over which variation rather than over whether the list is complete.

The idea

A first principle is a claim in the problem that isn't derived from anything else in it, something you'd still hold if every convention around it were withdrawn. Reasoning from first principles means separating those claims from the inherited ones stacked on top of them, and then building an answer up from the first group rather than editing the second.

It runs in two directions, and they're separate jobs. Going down, take each part of the current answer and ask what makes it so, then ask the same of that answer, and keep going until you reach something checkable or until the reply becomes "that's how it's done here", which is where an inherited claim was hiding. Going back up, build a new answer out of what survived, with the old one face down on the table, because editing the old answer reliably produces the old answer.

Enlarge: The descent finds the floor, and the ascent is a separate journey.

A column of four stacked labels descending from the inherited answer, to the convention behind it, to the decision behind that, to what is actually true, with the word why between each pair. A separate arrow runs from the bottom of that column up the right-hand side to a box reading a different answer, annotated rebuilt rather than edited.

The descent finds the floor, and the ascent is a separate journey.Close

The descent finds the floor, and the ascent is a separate journey.

The version everyone quotes is a cost estimate broken into its materials, where the assembled price of a component turns out to be several times the market price of the metal inside it, and that gap is the whole opportunity. It works because metals have public prices. Most management problems have nothing comparable, and the floor you reach for a hiring process or a pricing tier is a judgement somebody made rather than a number anyone can look up. So mark what you find as you find it. Facts hold whoever is in the room, contracts hold because somebody signed something and they carry a renewal date, and preferences hold because a person prefers them, which means the person can be asked.

How to use it

Pick a decision that's been stable, unsatisfying and expensive for a while, rather than whatever arrived this week, because the method costs hours and most decisions don't repay them.

Write the current answer down as one sentence, then write underneath it the reason it's that way, then the reason for that reason, and keep going for four or five steps. Somewhere in the descent the answers stop being reasons and start being history, and the line where that happens is the useful part of the whole exercise.

Sort what you're left with into the three kinds above, and be honest about which of them a thing really is, since a preference tends to describe itself as a constraint when somebody senior holds it.

Then rebuild forwards from the facts and the contracts, with the original out of sight, and give the rebuild a fixed amount of time, because generative work expands to fill whatever it's given.

Expect to arrive back where you started a good deal of the time. Convention is usually the accumulated result of other people having done this exercise, so agreeing with it is evidence that yours ran properly rather than that it was wasted.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

It's slow, and the slowness is why it mostly doesn't get used. A decision worth taking apart is load-bearing, repeated and hard to reverse, and a year holds a handful of those. Applied to everything, the method stops being a method and turns into a way of never settling anything.

The floor is softer than the engineering examples suggest. In physics the descent ends at something measurable. In an organisation it usually ends at somebody's preference, and a preference restated as a first principle is more dangerous than the assumption it replaced, because it has now been through a process and carries the authority of one.

It licenses ignoring things people learned the hard way. A constraint whose reason you can't find isn't the same as a constraint with no reason, and the approval step that looks like bureaucracy is sometimes scar tissue from an incident nobody wrote down. Chesterton's fence is the standing objection here and it's a good one, so ask who would know before concluding that no one does.

The evidence is anecdotal and the anecdotes are selected. The examples in circulation are the ones where somebody reasoned from the ground up and turned out to be right, because the people who did the same and were wrong don't get written up. Nothing here has been tested against a control group, and the honest claim for the method is that it widens the set of options rather than that it improves the hit rate.

It's socially expensive. Asking why something is done the way it's done reads as a challenge to whoever does it, however carefully it's phrased, and it reads that way hardest when the person can't answer. Run it on a decision you own before running it on somebody else's.

In one line

Take the answer apart until you reach something you could check, then build a new one without looking at the old one.