--- title: "What Is a Commitment Device? (And Do They Actually Work?)" description: "A commitment device is a choice you make now to constrain your future self. Here's the behavioural economics behind it, what the research actually shows, and when it backfires." source_url: "https://thecommitapp.com/commitment-device/" updated: 2026-08-12 --- # What Is a Commitment Device? (And Do They Actually Work?) ## The short definition A commitment device is an arrangement you enter into voluntarily, in a moment of clarity, that makes it harder or more expensive for your future self to do the thing you'll want to do but shouldn't. You are deliberately removing an option from someone you know will be weaker than you are right now. The canonical example is three thousand years old. Odysseus wanted to hear the Sirens, knew he'd steer toward them, and had his crew lash him to the mast and refuse his own orders. He didn't try harder. He changed what was possible. Notice what is doing the work there, because it isn't money. Odysseus paid nothing. The defining feature of a commitment device is simply that not keeping your word costs you something you agreed to in advance. Money is the most common form of that cost, because it is easy to measure and easy to move — not because it is the only one. The cost can be a blocked app, a set of pushups you owe before the screen time comes back, a task you take on tomorrow instead, or four specific people who will see that today is empty. What matters is that you set the cost while you were still being honest with yourself about how tomorrow tends to go. And the cost is only half of a working device. The other half is who decides whether you did it. A commitment you grade yourself is a wish with extra steps, so proof matters as much as consequences do: a serious consequence attached to a self-reported checkmark is only a more expensive way to lie to yourself. Both halves are covered below, in that order, because a device missing either one tends to fail quietly rather than loudly. ## Why anyone needs one Because preferences reverse. On Sunday night you genuinely want to run on Tuesday morning. On Tuesday morning you genuinely want to stay in bed. Neither is a lie — economists call this time-inconsistent preferences, and the version of you that decides is never the version that has to act. Most self-improvement advice is addressed to the wrong person. It tries to argue with Tuesday-you at 6am, who is not listening and has already reframed the run as optional. A commitment device works instead by making Sunday-you the one with power — and Sunday-you is thinking clearly, well-rested, and slightly appalled at what Tuesday-you keeps getting away with. Willpower asks you to win the same argument every morning. A commitment device asks you to win it once, and then removes the argument. ## The two types Hard commitments remove the option outright. Deleting the app. Giving your partner the console cable. Freezing the credit card in a block of ice. A pension you cannot draw down early. These work well, and they only work where the option is physically removable — which most of life isn't. Soft commitments leave the option available and attach a cost to taking it. You can skip the run. It will just block the apps you chose until you make it up, or cost you a set of squats to earn that screen time back, or show up as a gap on a board three people you named are looking at — or, in the version everyone thinks of first, cost you $20 you decided about on Sunday. Those are the same mechanism wearing different clothes. Most modern commitment devices are soft, because most goals can't be locked behind a physical barrier. Soft commitments have a failure mode worth naming: if the cost is too small, you'll simply pay it and feel briefly clever. The cost has to sit above what researchers informally call your motivation point — the level at which paying stops feeling like a reasonable trade. That threshold is personal, and it is not always financial. For some people the unbearable version is being seen not to have shown up. ## The half everyone skips: who checks Every self-reported tracker works until the streak is worth something. Then the checkmark stops being a record and quietly becomes a decision, and you are the person with the most to gain from a generous reading of what you did today. That is the whole argument for not being the one who reads it. So the second question of any commitment device, immediately after "what does a miss cost", is "who says whether I missed". There are three honest answers, in rough order of how hard each is to fake: - You do, honestly. Free, and fine for as long as it holds — which for most people is up to the first day when the honest answer is inconvenient. - Evidence does. A photo taken at the time rather than found in the camera roll, a GPS check-in, a screen-time reading, health data, a time-lapse of the middle instead of the finish. This isn't surveillance. It's that the record exists before you get a vote on what it says. - A person does. Someone you named while you were still serious, who can look at the evidence and say no. The usual failure here isn't dishonesty, it's attrition: the friend who agreed in January stops opening the app in March. Any device that rests on a witness needs a stated rule for what happens when the witness goes quiet. Verification isn't a refinement you add once the consequence is set. It's the thing that makes the consequence mean anything. Without proof, the lying doesn't stop, it just moves up a level: from skipping the run to reporting a run you didn't do. ## Being watched is itself a commitment device This one gets left out of most write-ups, probably because it's free. Telling four specific people what you promised to do this week is a complete commitment device with no money anywhere in it. Quitting in private is easy. Quitting in front of people who watched you sign up for it is a different experience, and the difference is the device. It works the same way the expensive versions do: it drags the cost forward in time. The benefit of a run is months away and statistical. Someone noticing you didn't show up arrives tonight, whether or not you currently feel like caring. Two conditions decide whether it does any work: - The people have to be specific. Announcing a goal to an audience achieves close to nothing; a board full of strangers does less than one friend who asks on Thursday. Your absence has to register with somebody who would actually notice it. - The group can't start excusing you. A group that treats a miss as fine becomes well-attended proof that missing is fine, which is worse than no group at all. Visibility also stacks cleanly onto every other kind of device. The same promise, made where three people can see whether you kept your word, is a stronger promise than the identical one made privately — and nobody has to be unkind for it to work, because an empty square is visible by itself. ## What the research actually shows Here is where most articles overclaim, so let's be precise. The best-known trial is Giné, Karlan and Zinman (2010), published in American Economic Journal: Applied Economics. Smokers in the Philippines were offered CARES: a savings account they paid into for six months, settled by a urine test for nicotine and cotinine. Pass and the money came back; fail and it went to charity. - 11% of those offered the product took it up. - Those offered it were 3 percentage points more likely to pass the six-month test than the control group. - The effect persisted in surprise tests at twelve months — suggesting real cessation, not test-gaming. Three percentage points is a real effect and a modest one. It is not "commitment devices triple your success rate", and anyone telling you otherwise is selling something. What makes the result striking is the durability: the behaviour held after the contract ended, which is exactly what most interventions fail to do. The broader literature — Thomas Schelling's work on self-command, Thaler and Sunstein's Nudge, Ian Ayres' Carrots and Sticks — converges on the same shape. Commitment devices reliably help some people, meaningfully, and the people they help most are the ones who already suspect they need one. Self-selection is the mechanism, not a flaw in it. ## When they fail - Nobody checks. The most common failure by a distance, and the one people notice last. A commitment you grade yourself is a wish, and every self-reported streak collapses at the same point: the moment the number matters, marking the day complete becomes irresistible. No consequence survives an unverified checkmark, because the checkmark is what triggers it. - The cost is too small. $1 is a receipt, not a deterrent. If you'd shrug at losing it, you will shrug at losing it. - The cost is too large. Over-doing it produces avoidance: people stop setting goals rather than risk a painful loss, or they quietly disengage from the whole system. A consequence should sting, not frighten. - Nobody was told. A device you kept to yourself has one enforcement point, and it's the same tired person who wanted to skip. Naming a witness costs nothing and roughly doubles what the rest of the design is doing. - The terms were vague. "Exercise more" cannot be adjudicated. "Run 5k on Monday, Wednesday and Friday before 7am" can — and vagueness always resolves in favour of the version of you who is looking for a way out. - Life happened and the system had no answer. A device with no grace for genuine illness or emergency doesn't build discipline — it builds resentment, and then abandonment. ## Designing one that works - Make it specific enough to adjudicate. A stranger should be able to look at the terms and say yes or no without asking you a question. - Decide what proof looks like before day one. Not on your first ambiguous evening, when you have an interest in the answer. Pick the kind of evidence where faking it would take more effort than just doing the thing: the time-lapse over the finish-line photo, the GPS track over your own recollection. - Name who verifies, and what happens if they go quiet. A person, or something automated, but decided in advance. A device whose referee can silently stop reviewing has no referee. - Tell somebody specific. Not an audience. One to ten people who know what you promised and would notice the gap. This is the cheapest half of the whole design and the half most people leave out. - Set the cost at your motivation point. The loss you'd be annoyed by but not frightened of — usually higher than your first guess. It doesn't have to be financial: blocked apps, earned-back screen time and a visible empty square all clear the bar for some people. - If money is the cost, decide where it goes while you're calm. Choosing the destination in advance is part of the device. A cause you actively dislike is the strongest version — StickK's data on anti-charities supports this, and it's their idea. - Build in grace. A small, finite number of excused days keeps a genuine emergency from destroying the whole structure — without making excuses free. ## Apps that do this A few products implement commitment devices directly, and the interesting differences are in the two halves above — how a day gets verified, and who can see it — more than in the size of the consequence: - StickK — the original, from the Yale economists who ran much of the research. A human referee you nominate verifies your self-report, and supporters can follow along. A forfeit can go to a friend, a charity or an anti-charity. Free, web-based. Compared with Commit → - Beeminder — graph-driven, with deep automatic data integrations, so what gets verified is imported data rather than proof of the act. Solo by design: nobody else is watching. The pledge escalates every time you derail, and Beeminder keeps the money. Compared with Commit → - Commit — ours. Proof is captured in the app rather than uploaded from your camera roll, and a day doesn't count until it's verified, either by the people you named or by Commit AI. Those people also see whether you showed up, in a group of up to ten. A miss can cost nothing but the streak, or block the apps you chose, or route a stake to a charity, an anti-charity or to Commit as a service fee — you pick which, before you sign. iOS, currently in public beta. ### Want to try one? Commit is in public beta on iOS and free while it is. There's no App Store listing yet. If you'd rather start today with something already shipping, StickK and Beeminder both work and we've linked them above in earnest. ## Sources - Giné, X., Karlan, D. & Zinman, J. (2010). Put Your Money Where Your Butt Is: A Commitment Contract for Smoking Cessation. American Economic Journal: Applied Economics, 2(4), 213–35. aeaweb.org - Ayres, I. Carrots and Sticks: Unlock the Power of Incentives to Get Things Done. - Thaler, R. & Sunstein, C. Nudge. - Schelling, T. — work on self-command and egonomics. ## Keep reading All writing · All comparisons