So far this course has treated “commitment” as one big lever: tie your hands, and you change what the other player expects. But commitments come in exactly two shapes. You can promise pain, or you can promise pleasure. A stick or a carrot. Everything else — escrow, hostages, doomsday machines, contracts — is just plumbing to make one of those two things believable.
This lesson does two jobs. First, it splits commitment into its two flavours: threats and promises, and shows that a promise has the same credibility problem as a threat (you will be tempted to weasel out of it). Second, it introduces Thomas Schelling’s single most useful distinction — deterrence versus compellence — the difference between stopping someone from starting and forcing them to act. Getting these two axes straight is what separates a strategist from someone who just yells “or else.”
Before you read — take a guess
Intuition check before we start: which of these is generally HARDER to pull off with a credible threat?
Threats and promises: the stick and the carrot
Both threats and promises are conditional commitments — statements of the form “IF you do X, THEN I will do Y.” They differ only in whether Y hurts or helps the other side.
- A threat is a commitment to punish the other side if they do (or fail to do) something. “If you cross this line, I retaliate.” The stick.
- A promise is a commitment to reward the other side if they do (or fail to do) something. “If you disarm, I lift the sanctions.” The carrot.
The analogy is exactly the donkey and the cart. To move the donkey you can beat it forward (threat) or dangle a carrot ahead of its nose (promise). Both change the donkey’s behaviour by attaching a consequence to it — one aversive, one attractive.
Here is the part everyone forgets: both are worthless unless credible. A threat you would never actually carry out is a bluff; a promise you would never actually honour is empty air. In the last lesson we saw that a threat is credible only when carrying it out is (or is made to be) in your interest at the moment of truth. Precisely the same test applies to a promise. The donkey ignores a beating it knows you’re too soft to deliver — and it ignores a carrot it suspects you’ll snatch away the instant it steps forward.
One template, two signs
Threat and promise are the same object with opposite sign. Both say “IF you do X THEN Y.” A threat sets Y = something painful; a promise sets Y = something pleasant. That’s why the credibility machinery from the last lesson — contracts, burned bridges, reputation, delegation, automaticity — works for both. You are always answering the same question: when the moment comes, will I actually do Y?
Pitfall: assuming your carrot doesn’t need to be believed
People instinctively treat threats as the thing that needs backing up and promises as self-evidently sincere — “I’m offering you something nice, why wouldn’t you trust me?” But an offer of something nice is exactly the offer you’re most tempted to withdraw once you’ve got what you wanted. The credibility burden on a promise is not smaller than on a threat. It’s the same burden, wearing a friendlier face.
Why promises need credibility too: the temptation to renege
Walk through the timing of a promise. “If you finish the job, I’ll pay you.” The other party does their part first — they perform. Now it’s your turn to pay. But look at your incentives at that instant: they’ve already delivered, the work is done, it can’t be undone. Paying costs you money and buys you nothing further. The purely self-interested move is to renege — to break the promise and keep the reward.
This is the promise-keeper’s version of the credibility problem, and it’s structurally identical to the empty threat. In both cases, after the other side moves, the payoff you dangled is no longer in your interest to deliver. And the other side, being no fool, sees this coming by backward induction (the mechanism from lesson 2): if they know you’ll stiff them at the end, they won’t perform at the start. The whole exchange collapses before it begins.
Real examples of the renege trap:
- “I’ll pay you Tuesday for a hamburger today.” Tuesday arrives; the burger is eaten; the incentive to pay has vanished.
- The contractor paid in full up front. With the money already in hand, finishing the job is now pure cost. Half-finished kitchens are monuments to promises that weren’t structured to survive.
- The government’s amnesty promise. “Lay down your arms and we won’t prosecute.” Once the rebels have disarmed and are harmless, the state’s temptation to prosecute anyway — to look tough, to satisfy victims — is enormous. Rebels who anticipate this never disarm.
Worked example: staging a promise so it survives
Suppose you hire a freelancer for a $4,000 project and promise to pay on delivery. The freelancer runs the same backward-induction check you would: “Once I hand over the files, what stops them from ghosting me?” Nothing — so they demand payment up front, at which point you face the mirror-image fear: pay first and they might vanish. Each side’s promise is incredible to the other. Deadlock.
The fix is the same toolkit that rescues threats — restructure the game so that keeping the promise is in your interest when the moment comes:
| Device | How it makes the promise credible |
|---|---|
| Escrow | A neutral third party holds the $4,000. Neither side can renege; the money is released on verified delivery. Both promises are now automatic. |
| Staged payments | Split into four $1,000 milestones. Each stage, the amount at stake is small, and both sides want the next stage to happen — so nobody’s tempted to defect on the final leg. |
| Hostage | You leave something of yours that you’ll lose if you renege — a deposit, a bond, a reputation stake. Reneging now costs you, so you won’t. |
| Reputation | If you’re a repeat player whose next ten clients are watching, stiffing this freelancer torches future business. The long game makes honouring the promise the selfish choice. |
Notice that every device does the same thing: it moves the cost of reneging forward so that, at the moment of truth, keeping the promise beats breaking it. That is credibility manufactured, exactly as for a threat.
A landlord tells a tenant: 'Renovate the apartment at your own expense and I'll extend your lease five years at the current rent.' The tenant refuses to renovate. What's the most likely strategic reason?
Deterrence versus compellence: Schelling’s key distinction
This is the heart of the lesson. Two threats can be worded almost identically and yet be worlds apart in how hard they are to make work — because they aim at opposite things relative to the status quo (the current state of affairs).
Deterrence is a threat designed to stop the other side from starting something — to preserve the status quo. “Invade, and we retaliate.” You are not asking anyone to do anything; you’re asking them to keep not doing it. Deterrence is passive: the threat just sits there, indefinitely, with no clock running. It has no deadline — “don’t ever invade” needs no Friday. And its success is invisible: when deterrence works, nothing happens. No invasion, no headlines, no medal. The dog that doesn’t bark is your triumph, and no one will thank you for it.
Compellence (Schelling’s coined term) is a threat designed to make the other side start doing something, or stop something they’re already doing — to change the status quo. “Withdraw your troops by Friday, or else.” Now you’re demanding action. Compellence is active: it must overcome inertia, not preserve it. It needs a deadline — without a “by Friday,” the target can comply “eventually,” i.e. never, and your threat is toothless, a permanent maybe that never forces the issue. And it demands visible compliance: the troops must actually be seen withdrawing, on your timetable, in a way everyone can verify.
Why compellence is so much harder
Three reasons, all flowing from that status-quo difference:
- Loss of face. Compellence requires the other side to visibly back down — to publicly reverse a course they had chosen. Backing down in front of an audience is humiliating; leaders will absorb real costs to avoid looking like they surrendered. Deterrence asks for no such humiliation: the target complies simply by not acting, quietly, and can always tell itself and its people it “was never going to invade anyway.” No face is lost when nothing visibly happens.
- The deadline forces a confrontation. A compellent threat’s clock creates a specific moment of showdown — comply by Friday or we clash Saturday. That collision is what you were hoping to avoid, and now you’ve scheduled it. A deterrent threat has no such date; it can succeed silently for decades with no showdown ever occurring.
- You must specify exactly what counts as compliance. With compellence you have to define the demand precisely — which troops, withdrawn to where, by when, verified how. Every ambiguity is a loophole the target exploits (“we did withdraw — those two units”). Deterrence needs no such specification: “don’t invade” is self-defining. Any invasion is a clear violation; there’s nothing to argue about.
The through-line: deterrence lets the other side save face by not-doing something quietly; compellence forces them to do something visibly, on a clock, against their pride. That’s why, other things equal, it is far easier to keep someone from starting than to make them stop.
Side by side
| Dimension | Deterrence | Compellence |
|---|---|---|
| Goal | Stop them from starting — preserve the status quo | Make them start (or stop) — change the status quo |
| Posture | Passive: the threat just sits there | Active: must overcome inertia |
| Deadline | None — it holds indefinitely | Required — no clock means it’s ignored |
| Who moves | Nobody, ideally (target keeps not-acting) | The target must visibly act / reverse course |
| What success looks like | Invisible — nothing happens | Visible compliance you can verify |
| Face | Target saves face (complies quietly) | Target loses face (must back down publicly) |
| Relative difficulty | Easier | Harder |
A memory hook
Deterrence = “Don’t you dare.” No deadline, invisible success, easy. Compellence = “Move it, and I mean by Friday.” Deadline, visible surrender, hard. If your demand has no clock and no specified action, you’re doing deterrence — and if you’re actually trying to change someone’s behaviour, that’s the wrong tool.
A regulator tells a factory: 'Cut your emissions below the legal limit within 90 days or we shut you down.' Classify this and name its most important design feature.
The 2×2: threat/promise crossed with deter/compel
The two axes are independent, so combining them gives four flavours of commitment. Every strategic move you’ll ever make lands in one of these boxes.
| Deter (stop them starting) | Compel (make them act/reverse) | |
|---|---|---|
| Threat (stick) | Threat-to-deter: “Invade us and we retaliate.” | Threat-to-compel: “Withdraw by Friday or we strike.” |
| Promise (carrot) | Promise-to-deter: “Stay out of the arms race and we’ll keep sharing this trade deal.” | Promise-to-compel: “Disarm by Friday and we lift all sanctions.” |
Read the row to know whether you’re using pain or pleasure; read the column to know whether you’re freezing the status quo or changing it. A promise-to-compel, note, still needs a deadline (carrots go stale too) and still needs to be credible against the renege temptation — you must actually lift those sanctions once they disarm, or next time no one will believe you.
Match the move to the model
Pair each term or scenario with what it is. Mind the traps — a “reward for stopping” is a promise-to-compel, and whether a move is deterrence or compellence depends on who currently owns the status quo.
Match each term or scenario to what it is.
Pick a term, then click its definition.
Pitfall: when the labels blur — and why misclassifying costs you
These categories are sharp in theory and slippery in practice. Two ways they blur, and one expensive consequence.
A “threat” to withhold a reward is really a promise. Suppose you’ve been paying a supplier a bonus, and you say: “Miss the deadline and you lose the bonus.” That sounds like a threat, but withholding a bonus isn’t inflicting pain — it’s declining to hand over a carrot. Whether a move is stick or carrot depends on the baseline you measure from. Relative to “bonus expected,” losing it feels like punishment; relative to “no bonus at all,” it’s just an unpaid promise. Frame the baseline and you flip the label. Strategically this matters because sticks and carrots have different credibility problems: you must worry about carrying out a threat, but about not reneging on a promise.
Deterrence vs compellence depends on who owns the status quo. “Get off my land” is compellence if the trespasser is already standing on it (you’re forcing a change), but “don’t step on my land” is deterrence if they haven’t crossed yet (you’re preserving the status quo) — same land, same words almost, opposite category. Whoever currently holds the ground defines which player has to move, and therefore which tool you’re really wielding.
The expensive consequence: using the wrong tool. The classic error is running a compellent threat with deterrent design — issuing a demand to change behaviour but attaching no deadline and no specified action. “You really should withdraw those troops” with no Friday and no definition of “withdraw” is a compellent goal dressed as a deterrent threat, and the target simply nods, stalls, and never moves — the threat is ignored precisely because it lacks the clock that compellence requires. Misclassify the move and you’ll reach for deterrence’s leave-it-open patience when the situation demands compellence’s deadline-and-specificity — or vice versa, slapping an aggressive ultimatum on a situation where quiet deterrence would have preserved everyone’s face.
Spot the trap: a CEO announces internally, 'Anyone still not using the new expense system should really switch over soon.' Employees keep using the old system for months. What went wrong, strategically?
Key takeaways
- Every commitment is a threat (a conditional promise of punishment — the stick) or a promise (a conditional promise of reward — the carrot). Both share the template “IF you do X, THEN Y,” and both are worthless unless credible.
- A promise faces the same credibility problem as a threat: after the other side performs, you’re tempted to renege. Anticipating this by backward induction, they won’t perform. Fix it the same way you fix a threat — escrow, staged payments, hostages, reputation — by moving the cost of breaking your word forward.
- Deterrence stops the other side from starting (preserve the status quo): passive, no deadline, invisible success, target saves face, easier.
- Compellence makes the other side start or reverse an action (change the status quo): active, needs a deadline, demands visible compliance, forces the target to lose face, harder.
- Compellence is harder because it requires a public back-down (loss of face), a deadline that forces the confrontation, and a precisely specified demand. Deterrence lets the target comply quietly by simply not-acting.
- Cross the two axes for four moves: threat-to-deter, threat-to-compel, promise-to-deter, promise-to-compel.
- The classic blunder is using the wrong tool: running a compellent goal with deterrent design — no deadline, no specified action — so the demand is ignored. Whether a move is a stick or a carrot, and whether it deters or compels, depends on the baseline and on who owns the status quo — read those before you choose your tool.