Commitment Devices

What is a commitment device?

A commitment device is any structure you set up in advance that constrains your future behaviour in a direction you currently endorse. The economist Thomas Schelling called this self-command. The act of one version of yourself binding another version. You agree to a deal at one moment that you cannot easily renegotiate at another.

The classic example is the alcoholic who pours every bottle in the house down the sink. Sober-you is making a contract with future-craving-you. The contract has teeth: if craving-you wants a drink at 11pm on a Tuesday, the only options are to put on a coat and walk to the off-license in the rain, or to honour the deal. The friction is the mechanism. Sometimes the friction is enough. Often it is not, which is why people graduate to commitment devices with real consequences.

The defining feature of a commitment device is that it costs future you something to break the contract. The cost can be financial (StickK takes your money and gives it to a charity you hate). It can be social (you have told ten friends, and breaking it means publicly admitting failure). It can be technical (the alcohol is gone, the gambling app is uninstalled, the iPhone is in a Yondr bag). The shape of the cost varies. The mechanism does not.

What a commitment device is not: it is not a notification, a reminder, or a habit tracker. Those tools depend on you choosing to engage with them in the moment of weakness. A commitment device works when you have already lost that argument, because the choice was made in advance.

Where the idea came from

The oldest commitment device in literature is in Homer's Odyssey. Ulysses wanted to hear the song of the Sirens but knew that any sailor who heard it would steer the ship into the rocks. So he had his crew tie him to the mast and stop their own ears with wax. Ulysses gave instructions in advance: do not untie me, no matter what I beg.

When the Sirens sang, Ulysses begged. The crew, deafened by wax, did not hear him. The ropes held. The ship sailed past. The Greeks called this akrasia. Weakness of the will. And they took it for granted that the rational mind in calm moments knew things the emotional mind in tempting moments would forget. The whole point of being tied to the mast is that you do not trust yourself to follow your own plan.

The phrase "Ulysses contract" entered modern medical ethics in the 1970s, applied originally to advance directives and then more broadly. Patients with bipolar disorder began signing agreements that authorised future psychiatric intervention even if they (in a manic state) refused it. The patient's stable self was overruling the patient's destabilised self. With consent given in advance. It is the same mechanism Homer described, applied to mental health.

In economics, Thomas Schelling published Egonomics, or the Art of Self-Management in 1980 and made the mechanism formal. Schelling argued that humans are not unitary actors. We are made of multiple selves with different time horizons and different preferences. The "self" who plans on Sunday is not the "self" who acts on Friday night. Commitment devices are how the planning self imposes its will on the acting self. The whole field of behavioural economics that followed. Kahneman, Tversky, Ainslie, Thaler. Rests on this foundation.

The modern, productised commitment device is a recent phenomenon. StickK launched in 2008, founded by the Yale economists Ian Ayres and Dean Karlan. It lets you stake money on a goal. Weight loss, smoking cessation, exercise. And lose the money to a charity (or, more potently, to an "anti-charity" you actively oppose) if you fail. Beeminder followed in 2011 with a more granular, data-driven version. Forfeit launched in the early 2020s with photo verification and a per-task stakes model. ScreenFine, launched in 2026, is the first to apply the mechanism specifically to screen time, with continuous, automatic monitoring rather than discrete per-task stakes.

The behavioural research behind it

Three findings from behavioural economics explain why commitment devices outperform good intentions.

Hyperbolic discounting

The economist George Ainslie showed in the 1970s that humans do not discount the future at a constant rate. We discount it hyperbolically: the value of a future reward drops sharply as it gets closer to now, then levels off. The practical result is that an outcome 10 years away and an outcome 10 years and one day away feel almost identical, but a reward today and a reward tomorrow feel hugely different. We are dramatically more impatient about the immediate future than the distant future.

This is why we plan to exercise tomorrow and then do not exercise tomorrow when tomorrow becomes today. The plan was made when "tomorrow" was abstract; the action is required when "tomorrow" is now. Commitment devices defeat hyperbolic discounting by making the cost of inaction concrete and immediate. The charge fires now, not in some distant future where you can rationalise it.

Loss aversion

Daniel Kahneman and Amos Tversky published Prospect Theory in 1979 and changed economics. They demonstrated experimentally that humans treat losses as roughly twice as motivating as equivalent gains. Losing $5 hurts about twice as much as finding $5 feels good.

For commitment devices, this is the load-bearing finding. A reward for good behaviour ("save 30 minutes by closing Instagram") is a gain frame, and gain frames are weak. A penalty for bad behaviour ("25 pushups for the next 15 minutes you stay in") is a loss frame, and loss frames are strong. This is why notifications and gentle nudges fail where verified-exercise commitment devices succeed: the brain is built to respond to losses more than gains, and notifications offer no loss.

The hot-cold empathy gap

The behavioural economist George Loewenstein documented that humans in a "cold" emotional state systematically underestimate how powerfully a "hot" state will affect their decisions. You promise yourself, calmly on Monday, that you will not check Instagram during the next work session. By Friday at 4pm, exhausted and frustrated, the urge to check is overwhelming and Monday-you's promise feels distant and academic.

Commitment devices work because they bind hot-state-you to cold-state-you's preferences. Cold-you signs the contract. Hot-you cannot break it without paying the cost. The empathy gap is the disease; the contract is the treatment.

The four flavours of commitment device

Not every commitment device is built the same way. They differ in what they put on the line. Some bite harder than others. Most people end up using a combination. The right shape depends on the goal.

1. Financial stakes

You put money on the line. Examples: StickK (anti-charity model), Forfeit (per-task stakes), ScreenFine (continuous fines for screen-time overage). The cost is dated, itemised, and shows up on a card statement. Loss aversion does the work.

Sub-flavour: anti-charity. Instead of losing money to a neutral party, you lose it to an organisation you actively hate. StickK pioneered this. The behaviour change is dramatic in studies because the loss is now aversive twice over. Losing the money, and funding something repugnant.

2. Social stakes

You publicly announce the goal and the consequence of failure is reputational. Examples: Beeminder's optional referee system, ScreenFine's Wall of Shame and partner mode, the millennia-old practice of telling friends "I'm going to do X." The cost is embarrassment, and for many people that is a heavier currency than money.

Social commitment devices work best when the audience is small, specific, and people whose opinion you actually care about. A tweet to 5,000 followers is weaker than a text to your three closest friends. Diffuse public commitment is mostly performative; focused interpersonal commitment is mostly real.

3. Technical stakes

You remove the option entirely or add hard friction. Examples: deleting an app, blocking a website at the DNS level, locking the phone in a Yondr bag, the Brick hardware product. The cost is the friction of undoing the technical state. You have to actively reverse the constraint to break the deal.

Technical commitments work best for binary, episodic behaviours (do not gamble, do not drink) and worst for continuous, calibrated ones (use Twitter for 30 minutes, not 3 hours). For the latter, the binary "blocked" or "unblocked" state mismatches the actual behavioural target.

4. Automated enforcement

A computer monitors and applies the consequence without you having to log it, prove it, or check it. Examples: ScreenFine reading screen-time data via Apple's FamilyControls API and applying fines automatically; Beeminder pulling data from Apple Health, Strava, RescueTime, etc. The cost is the consequence (financial, restrictive, social), but the enforcement is automated.

This is the youngest flavour and the one with the most upside. Older commitment devices required manual logging or a human referee. Automated enforcement removes the cheating surface area: there is no log to fudge, no witness to negotiate with. The data is the data.

Why commitment devices outperform willpower

The simplest framing: willpower is a finite resource that gets used up. Commitment devices substitute structure for willpower. You spend a small amount of willpower once, when you set up the device, and you do not have to spend it again every time the urge hits.

A more useful framing: in any moment of temptation, the rational and emotional minds are negotiating. Without a commitment device, both sides have full freedom to win the argument. The emotional mind argues for the immediate reward, the rational mind argues for the long-term goal, and whichever is louder wins. The negotiation is exhausting and the rational mind does not always win, especially when tired, hungry, lonely, or stressed.

A commitment device removes the rational mind from the argument. The deal is already signed; the rational mind no longer has to advocate for it. The emotional mind is now arguing not against the rational mind but against the contract. And the contract has a cost. "I'll just check Instagram for ten minutes" runs into "that costs 25 pushups before you can re-open Instagram." The argument shape is different. The cost is concrete and the timeline is short.

Empirically: in a meta-analysis published by Karlan and others, participants who used a financial commitment device for goal achievement (gym attendance, smoking cessation, weight loss) succeeded at meaningfully higher rates than control groups. Forfeit, which uses photo verification and per-task stakes, has reported a 94% success rate across more than 75,000 forfeits as of 2026 (per their published figures). The numbers are not magic, but the pattern is consistent: real consequences produce more behaviour change than reminders alone.

Where commitment devices fail

Commitment devices are powerful, not universal. Here are the cases where they do not work, or actively cause harm.

Mental health conditions where the urge is not a choice. Severe addiction, OCD, or anxiety disorders can produce behaviour that is not responsive to financial incentives in the way commitment-device theory assumes. Punishing yourself for symptoms of a clinical condition can deepen shame without changing behaviour. Treatment for those conditions is the better path; commitment devices may be a useful adjunct, but not a substitute. If your phone use feels compulsive in a way that no level of friction or fine moves, please talk to a clinician.

Goals where the right answer is unclear. Commitment devices require a target. If you do not know whether you want to spend less time on Instagram or just on news, a device that fines all phone use will mis-target. Spend a week observing your own usage before signing the contract. Then narrow the target.

Stakes that are too low. A $1 stake on a real habit will not move anyone. The stake has to be uncomfortable enough to register. The behavioural-economics rule of thumb is: pick a stake that, if you lost it tomorrow, would slightly ruin your week. Less than that is theatre. Much more than that and you are gambling.

Stakes that are too high. Conversely, life-disrupting stakes ($1,000 per slip, the family's holiday fund) tend to backfire. The fear of losing them produces avoidance behaviour. You stop engaging with the device altogether, often by uninstalling or cancelling. The mechanism only works if you stay inside it. ScreenFine\'s exercise-redemption design exists for this reason: even on a heavy day the cost is an extra workout, not a cleared-out bank account.

Goals that need flexibility. If the underlying behaviour is genuinely contextual ("I need to be on social media for my work two days a week"), a rigid commitment device will collide with reality and you will end up disabling it during the genuine need, then forgetting to re-enable it. Set up the device with rules that match real life: pause windows, specific carved-out apps, work-mode hours. ScreenFine's pause feature is built for this.

The novelty effect. Most commitment devices feel highly motivating in week one and considerably less motivating by week six. The brain adapts to the new normal. The defence is variety: rotate which villain persona delivers the consequence, change the redemption target, raise or lower the stakes as the baseline behaviour shifts. A static contract decays; a slightly evolving one stays alive.

Commitment devices for phone use and attention

Phone use is a particularly hard target for traditional commitment devices because the behaviour is continuous, not discrete. Going to the gym is something you either did or did not do. Using your phone is something you do for 4.3 hours today, 6.1 hours tomorrow, 2.8 hours next Tuesday. The success criterion is not binary; it is calibrated.

This continuous quality breaks the standard StickK / Forfeit pattern. You cannot reasonably stake $50 on "use Instagram less today" because there is no clean checkmark at the end. You can stake $50 on "log a 30-minute workout today," but for screen time, the underlying signal is too granular for one stake.

The mechanism that fits is continuous, automated enforcement with a daily cap. The cost accrues per unit of overage rather than per task. The cap exists because runaway daily costs would push the user to disable the device. The continuous monitoring matches the continuous behaviour.

This is the design ScreenFine implements. You set a daily limit. Every fifteen minutes you go past the limit locks the offending apps until verified pushups, steps, or mindful minutes. Each fine is pending for 1 week, redeemable by 1,000 steps, a logged workout, ten mindful minutes, twenty-five camera-counted pushups, or twenty-five squats. The lock holds until you complete the redemption. The consequence is the redemption itself, not a separate bill. Pause anytime in settings if real life requires it.

Continuous monitoring is what differentiates ScreenFine from a generic habit-stake app. Forfeit can fine you for failing to log a workout; ScreenFine locks your apps the moment you cross the limit, in real time, while the behaviour is still happening. The proximity matters: by the time a daily Forfeit task fails at midnight, the connection between behaviour and consequence has frayed. By the time a ScreenFine lock fires at the 91st minute, the connection is fresh.

See the full Screen Time Cost Index 2026 for what the math actually adds up to (the average American would pay $1,825 a year at the daily cap), and the comparison page for ScreenFine vs Forfeit for a closer look at how a continuous-monitoring device differs from a per-task stake.

Choosing the right commitment device app for you

A short decision tree of commitment device apps to consider, calibrated by what is actually working for the specific behaviour you want to change.

If your goal is discrete and verifiable (gym attendance, daily writing, language practice, smoking cessation): start with Forfeit or StickK. Photo verification handles the proof; per-task stakes match the behaviour.

If your goal is continuous and data-driven (calorie target, daily steps, sleep schedule): Beeminder is built for this. The yellow-brick-road model handles the continuous signal well; integrations with Apple Health, RescueTime, and other sources mean the data flows automatically.

If your goal is screen time specifically: start with friction-based tools (One Sec, Opal, ScreenZen, Apple Screen Time) and see whether they work for your case. They are cheaper and lighter. If after a month your usage has not meaningfully changed, the friction is not enough. You need a real consequence. ScreenFine is built for that case. See the alternatives roundup for the full landscape.

If you are not sure whether the underlying behaviour is a problem worth solving: it almost certainly is not. Commitment devices are uncomfortable by design. Spend a week observing the behaviour you are concerned about. If at the end of the week it still feels like something you want to change and you have not changed it, then a commitment device makes sense. If it feels mostly fine, save your money.

Further reading

• Schelling, T. (1980). The Intimate Contest for Self-Command. The Public Interest, 60.
• Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2).
• Ainslie, G. (1992). Picoeconomics: The Strategic Interaction of Successive Motivational States within the Person. Cambridge University Press.
• Ayres, I., & Karlan, D. (2008). The work that became StickK.
• Loewenstein, G. (2005). Hot-Cold Empathy Gaps and Medical Decision Making. Health Psychology, 24(4S).
• Thaler, R., & Sunstein, C. (2008). Nudge: Improving Decisions About Health, Wealth, and Happiness. Yale University Press.

Related guides