Would It Change Your Action?

We treat “gather more information” as an unconditional virtue. More data, another study, one more opinion, a quick check — surely it never hurts. But information costs time, money, and attention, and a startling amount of it changes nothing about what you were going to do anyway. This lesson gives you the single blade that cuts worthless information away from valuable information, usually in about ten seconds.

The core claim, stated crisply

Here is the whole model in one line. The value of information is the difference between two payoffs:

$\text{value} = (\text{payoff of your best decision WITH the information}) - (\text{payoff of your best decision WITHOUT it})$

That’s it. You compare how well you’d do if you knew the answer against how well you’d do acting on what you already know, and the gap is what the answer is worth.

Two properties fall straight out of that definition, and they do most of the work:

• It is never negative. Learning something can’t make your best decision worse, because you’re always free to ignore it. Worst case, the information is useless and you do exactly what you’d have done anyway — a gap of zero. (Real tests can cost money or time; that’s a separate subtraction. The information content itself is worth $\ge 0$.)
• It is exactly zero whenever every possible answer leads to the same action. If knowing the answer wouldn’t change your move, the “with” payoff equals the “without” payoff, and the gap collapses to nothing.

That second property is the whole game. Information is not valuable because it is true, interesting, precise, expensive, or hard to get. It is valuable only when it can change what you do. Analogy: a map is worth nothing to a traveler who is going to walk due north regardless of what the map shows. The map can be gorgeous, laminated, satellite-accurate — worthless, because it has no grip on the decision.

The would-it-change-my-action test

The model becomes a practical tool the moment you turn it into a two-step ritual. Call it the would-it-change-my-action test:

1. List the actions you are actually choosing between. Not the questions you’re curious about — the concrete moves. Operate or don’t. Ship Tuesday or Friday. Hire her or the other candidate.
2. For each possible answer the test could give, ask: what would I DO? Walk through the outcomes one at a time. If your action is the same across every answer, the test is worth zero. If some answer would flip you to a different action, the test might be worth gathering — you’ve found where its value lives.

Notice how mechanical this is. You’re not estimating probabilities or accuracy yet. You’re just checking whether any branch of the answer tree leads somewhere different.

Worked example: the surgeon and the scan

A patient arrives with symptoms that clearly require surgery. The surgeon is deciding: operate, or don’t operate. A colleague suggests one more scan.

Run the test on the operate-or-not decision. Suppose the scan comes back “severe” — the surgeon operates. Suppose it comes back “moderate” — the surgeon operates. Suppose “mild but present” — still operates, because the symptoms already cross the threshold. Every branch leads to the same action: operate. So for this decision, the scan is worth zero, no matter how sharp the imaging.

Here’s the subtlety that keeps this from being a cheap gotcha: value is always relative to a specific decision. The same scan might be enormously valuable for a different decision — how to operate, which approach, where to cut, what to have on standby. There the answers genuinely diverge, so there the scan earns its keep. The test doesn’t say “the scan is worthless.” It says “the scan is worthless for the operate-or-not question.” Always name the decision before you price the information.

Value comes only from averted mistakes

Why does changing your action matter so much? Because the only way information pays off is by steering you away from a decision you’d otherwise have gotten wrong. Information is a mistake-avoidance device. If you were already going to make the right move, there was no mistake to avert, and nothing to pay for.

This gives you a second, powerful shortcut: the value of information is capped by the cost of being wrong. The stakes set the ceiling. A test whose worst avertable error would cost you $50 is worth at most $50 — even if it is a perfect, infallible oracle. Perfect information about a trivial decision is a trivial amount of value.

Work a quick number. Suppose you’re about to buy a $50 phone case, and there’s some chance it doesn’t fit your model. Reading reviews might save you from a $50 mistake. That’s the entire prize. If reading reviews takes twenty minutes you’d value at more than $50, you’ve spent more gathering the information than the information could ever return. The mistake you’re insuring against is small, so the insurance is worth little — regardless of how thorough the reviews are.

Flip the stakes and the same logic scales up. Choosing a $400,000 house? Now an inspection that averts a foundation disaster is guarding a five-figure or six-figure mistake, so paying $600 for it is obviously worth it. Same kind of information — “here’s what’s actually true about the thing” — but the value tracks the stakes, not the accuracy.

Three quick portraits: worthless, worthless, valuable

Let’s build your pattern-recognition with three fast cases. Two are worthless. One looks identical but is genuinely valuable — because the decision is open.

Worthless #1 — the weather for a trip you’ll take rain or shine. You’ve booked a non-refundable cabin for the weekend with old friends you rarely see. You’re going whether it’s sunny or storming. Checking the five-day forecast for the go-or-cancel decision is worth zero: both answers lead to “go.” (Checking it to decide what to pack — that’s a different, open decision, and there the forecast earns its value. Name the decision.)

Worthless #2 — the second opinion you’ve pre-committed to overrule. You’ve resolved to get a particular treatment no matter what. A friend urges a second opinion. But you already know that if it disagrees, you’ll go ahead anyway, and if it agrees, you’ll go ahead too. Every branch: proceed. For the treat-or-not decision, the second opinion is worth zero — and, worse, gathering it can feel like diligence while changing nothing. Accurate, expensive, reassuring, and worth zero.

Valuable — the same second opinion, decision genuinely open. Now change one thing: you are truly undecided between the aggressive treatment and a watchful-waiting approach, and you’ve decided in advance that a clear second opinion would tip you. Now the branches diverge — “yes, treat” leads to treatment, “no, wait” leads to waiting. The information can move you, so it has value, and its value is bounded by how bad the wrong choice would be. Same test, same accuracy as case #2. The only thing that changed is whether an answer could alter your action.

That’s the punchline of the whole lesson in one comparison: it is never the information that is valuable or worthless in itself. It is the marriage of the information to an open decision.

The trap: mistaking “accurate” or “interesting” for “valuable”

Here’s the error that costs organizations millions and individuals whole afternoons: treating accuracy or interestingness as if they were the same thing as value.

They are not even on the same axis. Accuracy tells you how likely an answer is to be correct. Value tells you whether a correct answer would change what you do. A perfectly accurate answer to a question that can’t move you is worth exactly zero. An oracle that truthfully tells you a fact you’ll act identically on regardless is an expensive way to feel informed.

This is the disease of the vanity metric — the dashboard tile everyone glances at and no one acts on. A number can be precise, real-time, beautifully charted, and utterly decision-inert. The tell is that no possible reading of it would change a single decision: the metric goes up, you do nothing different; it goes down, you do nothing different. That number is decoration, not information. The same goes for the report nobody reads before deciding, the A/B test whose result won’t change the roadmap, the survey you’ll rationalize past.

The cure is not “collect less data.” It’s “collect data attached to a decision.” Before you build the dashboard, name the decision it will drive and the reading that would change it. If you can’t, you’re building a very accurate ornament.

Make it a habit

The model earns its keep when it becomes a reflex you run before gathering, not a post-mortem you run after. The ritual is almost embarrassingly simple:

Before you gather any data, write down: “What would I do if the answer is X? What would I do if it’s Y?” If the answers match, stop — you already have your decision. Don’t gather.

Writing it down matters more than it sounds. Held only in your head, “I should double-check” masquerades as prudence. On paper, “If the study says yes I ship, if it says no I ship” is visibly absurd, and you save yourself the study. The habit converts a vague itch for reassurance into an explicit, checkable claim about whether an answer could move you.

A caution so you don’t overprune: the test asks whether an answer could change your action, not whether it’s likely to. If there’s a genuine branch where you’d act differently — even a rare one — the information can carry value, and the next lessons will show you how to price that value precisely. The test kills the definitely-inert checks. It doesn’t kill the long-shot check that guards a genuine downside.

Where this goes next

You now have the qualitative blade: information is worth something only if some answer could change your action, and its value is capped by the stakes of getting that action wrong. That single test already retires a huge share of pointless data-gathering — the scans that won’t change the operation, the forecasts for trips you’ll take anyway, the dashboards nobody acts on.

But “greater than zero” is not a budget. Once a test passes this filter — once you’ve confirmed some answer really could flip you — the natural next question is: how much, exactly, is it worth? What’s the most you’d rationally pay for it? The next lesson puts a number on the ceiling: the expected value of perfect information (EVPI) — the value of a flawless oracle, which is the highest any real test could ever be worth. After that we’ll deal with noisy tests that give imperfect answers (EVSI), work out when a test is actually worth its price, and close with the pitfalls that trip up even careful decision-makers. The blade told you whether to look. Next we learn how much looking is worth.