The 30-second version
The idea
A shop's window gets smashed. The glazier gets paid to replace it, so people say the vandalism "boosted the economy." That's true as far as it goes — but the shopkeeper now can't spend that same money on the new shoes he was planning to buy. The shoemaker loses a sale nobody notices. Repairing damage only moves money around; it doesn't create it. The fallacy is counting the visible repair spending as a gain while ignoring the invisible spending it crowded out.
Where it comes from
French economist Frédéric Bastiat laid this out in 1850 in an essay usually translated as "What Is Seen and What Is Not Seen." He used the broken window as the simplest possible case of a much bigger habit: judging a decision only by its visible, immediate effects while the costs it quietly causes elsewhere go unmeasured and unmentioned.
Why it still matters
Every time someone argues that a storm, a war, a product recall, or a layoff round is "actually good for the economy" because of the repair spending, replacement spending, or severance spending it generates, they are usually committing this exact error. Learning to ask "what would that money or that effort have otherwise done?" is one of the cheapest ways to spot bad economic and business arguments before you act on them.
Mental Model Compass
A quick way to place this model next to a few others in the series, so you reach for the right tool. The horizontal axis is how obvious the cost or benefit is; the vertical axis is how far out in time the real effect shows up.
The Broken Window Fallacy sits in the hidden cost, immediate effect corner: the spending you can see happens right away; the spending it displaces is real but invisible unless you deliberately go looking for it.
- Opportunity Cost is the general tool. The Broken Window Fallacy is the specific trap of forgetting to apply it when something gets destroyed or disrupted.
- Sunk Cost is about money already spent; this model is about money about to be spent, and what else it could have done.
- Pareto Principle sits apart — it's about where effort concentrates, not about seen-versus-unseen effects.
Learn it, one layer at a time
A careless boy throws a rock through a shopkeeper's window. A crowd gathers. Someone says: "Well, every cloud has a silver lining — the glazier will get six francs for a new pane, and he'll spend that on something else, and so on. Isn't it lucky, in a way, that windows get broken?"
Bastiat calls this "the seen": six francs flow to the glazier. Everyone in the crowd can point to it.
Before the window broke, the shopkeeper had six francs and a good window. Now he has a good window and no six francs — because he had to spend them on a window he already had. He was planning to use those six francs for a new pair of shoes. The shoemaker never gets that order. No one in the crowd sees the shoes that don't get made, because they don't exist to be seen.
Bastiat's real target wasn't broken windows. It was the habit of judging any policy, purchase, or disruption only by what happens next — the visible, immediate, countable part — while ignoring what would have happened to the same money, time, or attention if the disruption hadn't occurred.
The test is always the same question, asked twice: "What do I see happening as a result of this?" and then, deliberately, "What would otherwise have happened with those same resources, that I don't get to see because it never happened?"
A news report says a factory fire "will actually help the local economy" because of all the rebuilding contracts it will generate. Using this model, what's the one question you'd ask the reporter?
What this looks like across different rooms
"The product recall created work for three teams." True — but every hour engineering spends re-tooling a faulty line is an hour not spent shipping the next feature, and every dollar of recall cost is a dollar not spent on R&D or hiring. The recall may still be the right call for safety or trust reasons; it is never a business win just because it generated activity.
Seen
Overtime pay, contractor invoices, a visibly busy ops team.Unseen
The roadmap item that slipped a quarter; the hire that got frozen to cover the cost."The hurricane will be good for the construction industry." Reporters often repeat this almost unchanged from town to town after every major storm. Rebuilding spending is real and visible on the local evening news. What's invisible is every other thing that money and that labor would have built, bought, or funded if the storm had never hit — plus the homes and infrastructure that simply don't exist anymore and have to be replaced just to get back to where the town already was.
Seen
Construction permits spike; contractors fully booked; GDP ticks up on rebuilding activity.Unseen
Savings drained, insurance premiums rising for years, vacations and purchases that never happened, net wealth that is lower than before the storm."Good thing my laptop died — now I get the new model." You do get a nicer laptop. What you don't get is whatever that replacement money was already earmarked for: the vacation, the emergency fund top-up, the debt payment. Treating a forced, unplanned purchase as a stroke of luck skips straight past the plan it just wrecked.
"This outage was great — it forced us to finally harden the system." Maybe true in hindsight. But the engineering hours spent firefighting and then hardening came from somewhere: the feature backlog, a different team's sprint, someone's weekend. Calling an outage a hidden blessing without naming what it displaced is the broken window fallacy wearing a hoodie. The same applies to "AI will create more jobs than it destroys" — it might, but the honest version of that claim has to show the jobs it destroys and the work it displaces, not just the new job titles that are easy to point to.
Worked application: evaluating a "this will boost the local economy" claim
- State the visible claim plainly. "The city's new $40M stadium will create 500 construction jobs and boost local spending."
- Name exactly what you can see. Construction payroll, contractor profits, a few seasons of concession and ticket spending near the stadium.
- Ask where the $40M would otherwise have gone. Tax revenue not spent on the stadium could have funded schools, gone back to residents as lower taxes (who would have spent or saved it themselves), or funded other infrastructure with longer-lasting returns.
- Ask what economic activity is simply moved, not created. Most "new" stadium spending is local residents redirecting entertainment dollars they were already going to spend elsewhere in town — on restaurants, bowling alleys, movie theaters — which now see less traffic.
- Separate the real argument from the fallacious one. There may be a genuine case for the stadium — civic pride, regional draw, tourism from outside the area that wouldn't otherwise arrive. That's a real, seen-and-unseen-aware argument. "It creates jobs," on its own, is not, because almost any use of $40M creates some jobs somewhere.
- Decide using the comparison, not the activity. The right question was never "does this create jobs?" It was always "does this create more net value than the next best use of the same $40M?" Make the decision on that comparison.
Common mistakes when applying this model
- ⚠️Treating it as proof that no repair or response is ever worth doing. The model doesn't say fixing the window was wrong — the window did need fixing. It only says don't count the repair spending as a bonus. The correct response to a real disaster can still be the right call; it's just not an economic gift.
- ⚠️Applying it as a blanket argument against all public spending or stimulus. The fallacy is about ignoring displaced spending when resources are already fully employed. It is weaker in a genuine recession with idle workers and idle factories, where the "unseen" alternative use of the money may really have been nothing at all — this is the core of the long-running debate between Bastiat-style and Keynesian economists, and a careful user of this model acknowledges that debate rather than treating the fallacy as a universal trump card.
- ⚠️Confusing it with "Broken Windows Theory." Despite the nearly identical name, that is an unrelated idea from policing and urban studies about how visible small disorder (like actual broken windows left unrepaired) can signal neglect and invite more serious crime. The two concepts share a prop, not a claim — don't let the name overlap convince you they're the same model.
- ⚠️Using it to win an argument instead of to find the truth. It's tempting to deploy "that's just the broken window fallacy" as a conversation-ending mic drop. Used well, it's an invitation to go find the specific unseen cost, not a label that lets you stop thinking.
When not to reach for this model
This model is built for cases where resources are scarce and roughly fully used — the normal condition of a household, a team's time, or a healthy economy. It gets shakier, and needs a companion model, in a few situations:
- 🛑Deep recession or genuinely idle resources. If the money, workers, or factory capacity would otherwise have sat completely unused, the "unseen alternative" may be smaller than the model assumes. Pair this with basic macroeconomic context before using it to argue against relief spending.
- 🛑Safety, risk, and insurance decisions. "Should we repair the faulty brake line" is not an economics-of-stimulus question. Don't let seen-versus-unseen reasoning talk you out of a repair that's actually about preventing harm.
- 🛑Decisions with real option value. Sometimes a disruption forces a genuinely overdue upgrade (the old system was going to need replacing anyway, soon, at similar cost). In that narrow case the "unseen cost" is smaller than it first appears, because the spending was only pulled forward in time, not newly created. Check the counterfactual honestly rather than assuming the fallacy always applies at full strength.
Try it yourself
Describe a real claim or decision you're weighing, and build a ready-to-use prompt that walks an AI assistant — or your own thinking — through the seen-versus-unseen check.
Save as a skill
This packages the model into a short, reusable instruction set you can hand to an AI assistant any time you want it to run the seen-versus-unseen check on something you bring it — no need to re-explain the model each time.