Sunk Cost Prison: Why Users Can’t Leave Apps They Already Hate

The sunk cost fallacy is why smart people keep pouring money into projects they secretly wish had never started, why gamers sink another $200 into a banner pull after the first $600 disappeared, and why entire governments stayed in wars for a decade longer than any rational planner would have allowed. The cost is gone. The decision is about what happens next. Every serious behavioral scientist since Richard Thaler first named the bias in 1980 has agreed on that. And almost nobody behaves that way.

A teenager keeps a Snapchat streak alive at 2 a.m. with a friend she stopped actually talking to six months ago. A Duolingo user cries when a multi-year streak breaks. A parent keeps scrolling Facebook even though nothing on the feed has made her smile in a year. None of these people are stupid. All of them are caught by the same 45-year-old cognitive bias, now weaponized into interface design.

This is the full expert guide to the sunk cost fallacy — the classic experiments that proved it, the neuroscience that explains it, the places where the bias breaks down, and (because I am a behavioral designer, not just a behavioral scientist) the ethical line between using sunk cost to help users finish what they started and using it to trap them inside products they have started to hate.

Speed Run Notes

• The sunk cost fallacy is the tendency to continue investing in a losing course of action because of what has already been spent, rather than what can still be gained. The money, time, and effort are gone either way.
• Richard Thaler named it in “Toward a Positive Theory of Consumer Choice” (1980). Arkes and Blumer’s 1985 theater-ticket experiment is still the canonical proof: people who paid $15 for a season ticket attended significantly more plays than people who got the same ticket for free.
• The underlying machinery is loss aversion (Kahneman & Tversky, 1979). Losses feel roughly twice as painful as equivalent gains feel pleasant, so “wasting” a past investment hurts more than the marginal cost of continuing.
• Escalation of commitment (Staw, 1976) is sunk cost on the organizational scale: the Concorde supersonic jet, the Vietnam War, and dozens of failed enterprise software projects kept going for the same reason a gambler raises their bet to chase losses.
• The fallacy is weaker in children, older adults, and non-human animals (Arkes & Ayton, 1999; Strough et al., 2008). A four-year-old, a pigeon, and a 75-year-old are each less likely to throw good resources after bad than a thirty-year-old MBA.
• Neuroscience links sunk cost to the anterior insular cortex and anterior cingulate cortex: the same regions that light up for physical pain and social rejection. The bias is literally an emotional pain response to waste.
• In the Octalysis Framework, the sunk cost fallacy lives inside Core Drive 8: Loss & Avoidance, and its designed form is Game Technique #50: the Sunk Cost Prison. Duolingo streaks, gacha banners, and Facebook social graphs are the canonical modern examples.
• The counter-move is not logic. The counter-move is FOMO Punch (Game Technique #84). Fear of missing a future opportunity reliably beats fear of losing a past investment: which is how almost every successful migration off a legacy platform actually happens.
• The elephant in the room: every time you stay in a product, relationship, job, or investment mainly because you have already invested, you are handing your future to your past self. The ethical designer’s job is to notice when their product is asking users to do that.

About Yu-kai Chou

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and Google Scholar — with 3,700+ more academic publications. Explore his books here.

What Is the Sunk Cost Fallacy?

The sunk cost fallacy is the tendency to continue investing in a course of action because of resources already committed to it, rather than because of the resources that are still at stake and the value that can still be gained. The sunk cost — the money, time, effort, or opportunity already spent — cannot be recovered by any future decision. Rational choice theory therefore says to ignore it. Humans ignore almost everything else instead.

The bias was first formally named by economist Richard Thaler in his 1980 paper “Toward a Positive Theory of Consumer Choice”. Thaler was, at the time, attempting to catalogue the ways real consumer behavior diverged from the utility-maximizing agents in textbook microeconomics. He described a hypothetical man who had paid $40 for tickets to a basketball game and then, when a blizzard made attendance miserable, insisted on driving anyway. The $40 was sunk. The question was whether driving through the blizzard to watch the game was worth more than staying home. The sunk cost had no bearing on that question, but it fully drove the man’s decision.

Thaler embedded the observation inside a broader project: mental accounting. Humans, he argued, do not hold a single pool of wealth and compute expected utility against it. They keep many small mental ledgers — this is the movie-tickets budget, that is the vacation budget, the other is the business investment — and they want each ledger to close in the black. A sunk cost is a ledger that has already gone into the red. Continuing to invest feels, from inside the mental-accounting frame, like the only way to save the ledger. Walking away confirms the loss.

Five years after Thaler’s paper, two psychologists at Ohio University published the experiment that turned the theoretical observation into an unambiguous empirical fact.

The Three Classic Experiments

1. Arkes and Blumer: The Ohio Theater Season Ticket (1985)

In 1985, Hal Arkes and Catherine Blumer ran a field experiment at the Ohio University Theater. Customers arrived to buy season passes for a university theater series. Without the customers’ knowledge, the researchers randomly assigned them to one of three ticket prices: the full $15 list price, a $2 discount to $13, or a $7 discount to $8. The customers paid what their assigned sticker said. Everyone received the same season pass, to the same plays, in the same theater, from the same seats.

Over the course of the season, Arkes and Blumer counted how many plays each group actually attended. By every rational theory of choice, price paid at the front door should have no effect on attendance once the season had started: the money was gone either way, the expected enjoyment of each play was identical. But customers who had paid full price attended significantly more plays than customers who had been given a discount. The effect was so strong that by the end of the first half of the season, full-price patrons were showing up an average of 44% more often than the most heavily discounted group. In the second half of the season the effect faded, as the sunk-cost memory grew less vivid.

This is the canonical proof that the sunk cost fallacy is not a story economists tell themselves. People really do spend more time consuming something because they once spent more money on it, even when the money has no bearing on the experience in front of them.

2. Garland: The Oil Well Scenario (1990)

In a 1990 study, Howard Garland (University of Delaware) gave business students a simulated decision. They were managers of an oil exploration project that had already cost a portion of its projected budget. They were asked whether to continue the project or abandon it, given a stated low probability of success. The key manipulation was how much of the total budget had already been spent: 10%, 30%, 50%, 70%, or 90%.

The rational answer depended only on the forward-looking expected value of the remaining spend, not on what was already gone. But Garland’s subjects got steadily more willing to continue as the sunk percentage grew. A project that was 90% funded and likely to fail still attracted enthusiastic commitment to finish, while the same project with only 10% spent was abandoned with a shrug. Thirty years later, this is still the cleanest demonstration that sunk cost scales with accumulated investment: the more you have poured in, the harder it gets to walk away, even though the pouring-in is precisely the part of the problem that no longer depends on you.

3. Staw: Knee-Deep in the Big Muddy (1976)

The third classic is slightly older than the other two and bigger in ambition. In 1976, organizational behaviorist Barry Staw published “Knee-Deep in the Big Muddy: A Study of Escalating Commitment to a Chosen Course of Action”. Staw gave business-school students a case study in which they, as newly minted R&D executives, had allocated a research budget across two divisions of a fictional company and then, years later, had to decide whether to double down on whichever division was now underperforming.

Staw’s finding was that executives who had personally made the original allocation invested more additional money in the losing division than executives who were handed the same situation fresh. The difference was not trivial: it ran roughly 20% higher. Staw coined the phrase escalation of commitment to describe this specifically social version of the fallacy: the cost is not only financial, it is reputational. Pulling the plug on the losing project means admitting that the earlier allocation was wrong. Escalating is easier than explaining.

Together, these three papers — Arkes and Blumer on individual consumers, Garland on project managers, and Staw on organizational decision-makers — provide a base of evidence wide enough that no serious behavioral scientist disputes the existence of the effect. What they dispute, as we will see, is when it actually matters.

Escalation of Commitment: Sunk Cost at the Organizational Scale

Escalation of commitment is the operating temperature at which the sunk cost fallacy burns hottest, because organizations layer two additional forces on top of the base bias.

The first additional force is self-justification. The executive who authorized the original budget is usually the same executive deciding whether to continue it. Admitting that the first decision was wrong is, socially, identical to admitting that they were wrong. This is the same pressure that keeps a partner in a failing relationship past the moment they knew, or keeps a founder insisting the product-market-fit is “just around the corner” three pivots later.

The second additional force is audience effects. Joel Brockner and others documented in the late 1980s and early 1990s that the presence of an observing audience — a boss, a board, a public shareholder base — sharply increases the rate of escalation. It is much easier to write off a loss privately than to announce it in a quarterly call. When the organization is publicly accountable, the sunk cost calculus tilts even further toward continuation.

The two most-cited real-world examples of escalation are both national in scale. The first is the Concorde supersonic airliner, a joint British-French project whose development budget ballooned from an original £70 million estimate in 1962 to over £1.1 billion by its 1976 launch (a more-than-tenfold overrun that both governments knew was economically indefensible by the late 1960s), but which neither could abandon without admitting the money already spent had been wasted. The pattern became so synonymous with the bias that European economists still call it the Concorde fallacy.

The second is the United States military escalation in Vietnam, where each successive administration from Kennedy through Nixon faced a version of the same question: given the American lives and dollars already spent, is it acceptable to withdraw? The answer, for more than a decade, was no — even though, by any forward-looking calculation, the probability of the stated objectives was declining. Declassified cabinet memos from the period explicitly discuss the inability to “lose what we have already invested,” which is not a strategic argument. It is sunk cost running a superpower.

At a smaller scale, every failed enterprise software rollout, every overrun infrastructure project, and every biotech clinical trial that stays in Phase III past the point of clinical plausibility is running the same script. The technical term escalation of commitment has crept into plain English precisely because the pattern repeats at every organizational size, from a two-person startup to a continental war.

What Thaler and Arkes Got Right

Four pieces of the original sunk-cost story have held up cleanly across four decades of follow-up research.

First, the core claim was correct. People really do treat past expenditures as if they still matter, even when the rational choice would ignore them. Every large-sample replication — from business-school students to retail consumers to senior managers — produces an effect in the predicted direction. The Arkes-Blumer 1985 result has been replicated in theater subscriptions, gym memberships, ski-lift tickets, airline bookings, and prepaid mobile plans. The bias is real and it is robust.

Second, the effect is larger for self-generated than for externally imposed expenditures. A person who personally chose to spend $600 on a non-refundable concert ticket feels the sunk-cost pull more strongly than a person whose employer prepaid the same ticket as part of a package. The bias is tightly coupled to personal responsibility, which is part of why escalation of commitment spikes in founders, executives, and anyone with a public track record to defend.

Third, the mental-accounting framing survives. Thaler’s larger theoretical move — that humans keep many small ledgers rather than one big one — continues to explain anomalies that pure expected-utility theory cannot. The reason a concert-goer does not simply treat the non-refundable ticket as part of lifetime consumption smoothing is that they do not have a lifetime consumption account in their head. They have a “this ticket I bought” account, and it wants to close in the black.

Fourth, sunk cost predicts continuation, not always bad outcomes. A nuance often lost in popular summaries is that the sunk cost effect biases continuation, not failure. Continuation is sometimes the right call — the project really is 90% complete, the relationship really will improve, the language streak really is worth protecting. Thaler himself has been careful on this point. The bias is a bias, not a guarantee of a bad decision. A designer or manager who treats every case of persistence as irrational will destroy a lot of real value in the name of avoiding a bias.

These are the load-bearing pieces. They have paid for themselves many times over in both academic citation counts and dollars saved in project-management courses. The pieces that did not hold up are equally important, and much less often discussed.

Where the Sunk Cost Fallacy Falls Apart

Critique 1: Animals and Children Don’t Do It

In a widely cited 1999 Psychological Bulletin review, Hal Arkes and Peter Ayton asked a question that the original Arkes-Blumer paper had not: if the sunk cost fallacy is a deep feature of how minds evaluate resources, why don’t animals show it? The literature at the time had searched for sunk-cost behavior in pigeons, rats, monkeys, and even starlings. In foraging experiments where an animal had invested time or energy in a food source before a better patch appeared, every species tested abandoned the old patch as soon as expected returns favored the new one. No sunk cost.

Young children are similar. Arkes and Ayton summarized experiments with preschoolers and early-elementary children: given tasks where continuing was rational only if previous effort mattered, the children happily abandoned prior effort as soon as a better option appeared. Adult-grade sunk-cost thinking seems to develop somewhere between ages 10 and 14, around the same time children acquire a fully-formed sense of personal responsibility and reputation. That developmental profile is a very strong hint that the fallacy is not a low-level cognitive defect. It is a side-effect of becoming a social adult.

Critique 2: Laboratory Null Results Are More Common Than Textbooks Admit

In 2007, Daniel Friedman and colleagues (UC Santa Cruz) published “Searching for the Sunk Cost Fallacy”, a paper with a deliberately pointed title. They ran a series of tightly controlled laboratory experiments in which undergraduates made repeated decisions under varying sunk-cost manipulations, with real money stakes. Their finding was unfashionable: in clean experimental paradigms with experienced participants, the sunk-cost effect shrank, wobbled, and often disappeared altogether. They estimated that the commonly cited effect sizes from field studies overstate the bias in tightly controlled settings by a factor of two to five.

This does not disprove the effect: the Arkes-Blumer theater result was a field study for a reason, and field environments are where the bias actually operates. But it means the “universal irrational bias” framing of popular books is too strong. The fallacy is context-sensitive. It reliably appears in high-stakes, self-responsible, publicly observable decisions. It disappears in anonymous lab trials with instant feedback, which is why economists spent decades not finding it.

Critique 3: Older Adults Are Partially Immune

In a striking 2008 study, JoNell Strough and collaborators at West Virginia University found that older adults (ages 60–80) were significantly less susceptible to the sunk cost fallacy than young adults (18–30) in scenarios involving financial, health, and relationship decisions. Follow-up work has replicated this pattern across cultures. Strough’s team argues that older adults have both more experience with failed projects — they have seen what happens when you try to finish things that should be abandoned — and a different emotional relationship to the idea of waste. Time is more obviously finite at 75 than at 30, which makes continuing a losing course feel more clearly like spending what you cannot get back.

This matters because it places the fallacy where behavioral science is slowly re-placing most of the so-called universal biases — not as a fixed property of human cognition, but as a particular configuration of motivations that can be trained, softened, or even reversed with experience. A behavioral designer who treats sunk cost as a permanent design primitive, immune to the user’s own learning, is going to be surprised when their most valuable users — the long-tenured, the experienced, the loyal — suddenly walk out.

What’s Really Happening Inside the Brain

The brain systems that produce sunk-cost behavior are the same ones that process physical pain, social rejection, and anticipated loss. They are older than deliberative reasoning and faster than it.

Functional imaging studies consistently light up two regions during sunk-cost decisions. The first is the anterior insular cortex, a deep region involved in interoceptive awareness: the feeling of the body’s internal state. The insula activates for physical pain, for social rejection, for disgust, and for the feeling of being cheated. When a subject considers abandoning a project they have already invested in, the insula lights up the same way it does when they consider letting down a friend. Losing a sunk investment, at the neural level, is filed under the category of “a thing that hurts.”

The second region is the anterior cingulate cortex (ACC), which monitors conflict between decisions and tracks the emotional weight of errors. ACC activation correlates with the subjective feeling of waste, and ACC damage sharply reduces sunk-cost behavior. A 2018 study by Brian Sweis and colleagues at the University of Minnesota extended this work to rats and mice in a “restaurant row” paradigm where animals chose how long to wait for food rewards. Once a rat had waited long enough, the probability of continuing to wait increased sharply — an exact sunk-cost signature — and this behavior was driven by the rodent homolog of the ACC. Turning off that region eliminated the effect. So the claim earlier that “animals don’t do sunk cost” needs qualifying: the original foraging experiments were too coarse to catch it, and more fine-grained paradigms reveal a partial sunk-cost bias in rats and mice as well. The simple version of the developmental story is wrong, but the neural story is consistent: sunk-cost behavior is an ACC-plus-insula phenomenon.

Stress amplifies the bias. Anna Haller (Haller & Schwabe, 2014, NeuroImage) showed that participants placed under acute laboratory stress — the Trier Social Stress Test — made significantly more sunk-cost-consistent choices than unstressed controls. This is important for designers: the user who is most vulnerable to sunk-cost trapping is not the bored user who is browsing with a full battery and a free afternoon. It is the stressed user trying to maintain a streak on a bad day.

Finally, dopamine plays a role through Brian Knutson’s work on anticipatory reward processing. Dopamine neurons fire not at reward receipt but at reward prediction. Once a user has been predicting a payoff for a long time — a character pull, a streak milestone, a promotion — the dopamine system has built a forecast. Abandoning the course forces the brain to downgrade the forecast, and dopamine drops. Subjectively, that downgrade feels like a loss.

The picture that emerges from two decades of neuroscience is coherent: the sunk cost fallacy is not a cognitive bug in the deliberative system. It is the deliberative system being overruled, correctly and reliably, by an older network that represents waste, loss, and abandoned effort as physically painful events.

Sunk Cost vs. Other Behavioral Theories

Sunk Cost Fallacy vs. Loss Aversion

Loss aversion: Kahneman and Tversky’s 1979 observation that losses feel roughly twice as painful as equivalent gains feel pleasant: is the engine beneath the sunk cost fallacy, but it is not the same thing. Loss aversion is about how the brain weights a prospective change; sunk cost is about how the brain weights a completed one. A pure loss-averse agent would still ignore sunk costs, because the sunk money is gone either way: no prospective loss remains. What the sunk cost fallacy adds on top of loss aversion is the refusal to re-code a past expenditure as already lost. Most sunk-cost papers now agree that loss aversion is necessary but not sufficient. Mental accounting, self-justification, and audience effects do the rest.

Sunk Cost Fallacy vs. Commitment and Consistency (Cialdini)

Robert Cialdini’s commitment-and-consistency principle describes the human tendency to align future behavior with past stated positions: once you have publicly committed to a position, you will push that position further than you would have if asked fresh. Commitment-and-consistency and sunk cost both produce persistence, and they frequently reinforce each other, but they are driven by different engines. Commitment-and-consistency is social: it is the desire to look coherent to other people. Sunk cost is internal: it is the desire to not feel that your own past spending was wasted. In practice, the two operate together whenever a user has both publicly committed to a product (posted about their streak, told friends about a gacha pull) and spent real resources on it. That joint activation is what makes brand loyalty so sticky.

Sunk Cost Fallacy vs. The Endowment Effect

The endowment effect, formalized by Thaler in the same 1980 paper that named the sunk cost fallacy, describes the overvaluation of things we already own. The two biases share a common ancestor (loss aversion) and frequently bundle together in design, but they are distinct. The endowment effect operates on stocks: a mug you already hold, a subscription you already have, a relationship you are already in. The sunk cost fallacy operates on flows: money you have already spent, time you have already committed, effort you have already poured out. A product that is strong on both wins twice: users overvalue what they already have in the platform (endowment) and discount the mounting cost of continuing to use it (sunk cost). Facebook is the textbook case.

Sunk Cost Fallacy vs. Status Quo Bias

Status quo bias: the tendency to keep current arrangements unchanged: overlaps with sunk cost but is subtly different. Status quo bias applies even when no prior investment has been made; it is the cost of re-deciding. Sunk cost adds a loss-coded accounting layer on top. In product design, status quo bias is what keeps a user on the default plan. Sunk cost is what keeps them on the plan they chose after considering all the alternatives. The two stack: the user first chose your plan (sunk cost), then stopped re-evaluating (status quo). Most retention funnels run on this stack without naming it.

The Sunk Cost Fallacy in the Real World

In Business Strategy and Project Management

Every strategy consultant running a post-mortem on a failed product launch encounters the same pattern. Eighteen months in, the leadership team knew the product was not going to win. Twenty-four months in, they still could not kill it, because killing it meant writing off a year of engineering, a marketing campaign, a signed distribution deal, and a founder’s promise to the board. The project kept drawing resources for another nine months, often doubling its losses, before being quietly wound down.

The tactical fix that actually works is the pre-mortem: an exercise introduced by Gary Klein in which, at the start of a project, the team explicitly writes down the conditions under which they would kill it. Pre-registering the kill criteria removes the sunk-cost step from the future decision: the decision to abandon is not “we wasted $5 million,” it is “the $5-million threshold we agreed on was crossed.” Companies that pre-mortem seriously cut their escalation losses by roughly 30–40% in Klein’s case studies.

In Personal Finance and Investing

The individual investor’s version of escalation of commitment is holding a losing stock past the point of any rational thesis. Behavioral-finance researcher Terrance Odean showed in a landmark 1998 study of 10,000 brokerage accounts that retail investors sell winning positions 1.7 times more often than losing positions, even though tax optimization suggests the reverse. The reason is not stupidity. It is that selling a loser forces acknowledgment of the sunk cost. Holding the loser keeps the loss “unrealized”: a tellingly loaded word: and keeps the mental account from closing in the red. Odean estimated that this bias costs the average retail investor around 3.4% in annualized returns. A lifetime of that compounds to a house.

In Healthcare and Treatment Decisions

Physicians and patients both show strong sunk-cost effects in treatment decisions. Bornstein and Chapman demonstrated that the length of time a patient had already been on a treatment predicted willingness to continue the treatment more strongly than the treatment’s current efficacy. A cancer patient who has endured a year of chemotherapy with poor response is more likely to continue than a patient with the same prognosis who has only just started: because the first patient is also weighing a year of suffering as a sunk cost. Good clinical decision support, of the sort being rolled out in progressive oncology centers, explicitly reframes these decisions forward-only, removing prior treatment duration from the displayed decision surface.

In Relationships and Life Choices

The most painful domain. People stay in romantic relationships, friendships, careers, and cities long past the point of personal flourishing, often because they have already invested a decade. The forward-looking question: is this life, from today on, the best life available to me?: gets overwritten by the backward-looking one: would leaving mean the last ten years were wasted? The answer to the second question is always yes. The answer to the first is frequently no. Anyone who has sat with a friend considering a late-stage divorce or a mid-career pivot has watched the two questions fight each other in real time.

The Elephant in the Room

Here is the part of the sunk cost conversation that academic papers rarely say out loud. Most of the real sunk-cost decisions in our lives are not about money. They are about identity.

When a 35-year-old programmer refuses to leave the career she has come to dislike, the sunk cost is not the college tuition. It is the story she has told herself about who she is. Leaving means reclassifying ten years of self-description as a mistake. The money is trivial. The identity rewrite is the real cost. When the cabinet member insists on staying the Vietnam course, the sunk cost is not the billions spent. It is the claim that he has been right this whole time. Writing off the war means writing off the public image of himself as a wise strategist.

Every advanced design for getting users out of sunk-cost traps: the ones that actually work: addresses identity, not math. The successful migration off a legacy enterprise platform reframes the move as an upgrade, not an abandonment: the team is not admitting their last decade was wrong, they are graduating. The successful exit from a dysfunctional relationship reframes leaving as being true to a deeper, prior self, not as throwing away the years. The migration pitch that says “you were right to invest in X, and the skill you built is exactly what gets you to Y” converts orders of magnitude better than the pitch that says “X was a waste, come to Y.”

This is why a purely logical argument: “the money is gone either way”: almost never moves a person locked into a sunk cost. The money was never the problem. What is really on the line is whether the story of their past decisions is allowed to be a good story. The designer’s and the therapist’s job is the same here: protect the story while changing the direction.

How to Apply the Sunk Cost Fallacy with the Octalysis Framework

In the Octalysis Framework, the sunk cost fallacy lives inside Core Drive 8: Loss & Avoidance. It is Black Hat motivation: users comply because continuing feels less painful than walking away, not because the experience is intrinsically rewarding. CD8 pushes hard and produces fast behavior, and it also poisons the user’s long-term relationship with the product if it is the only driver of engagement. The ethical designer’s job is to use it carefully, with exits, and always paired with at least one White Hat Core Drive.

Game Technique #50: The Sunk Cost Prison

The designed application of the sunk cost fallacy is what I call Game Technique #50: the Sunk Cost Prison. It is a deliberately constructed state where users have invested so much time, money, effort, data, or social connection into a product that leaving feels psychologically unbearable. The word prison is deliberate. It is not just that the user can leave and chooses not to. It is that leaving feels like an act of destruction, like setting fire to something you personally built.

The most visceral modern example is the Duolingo streak. A streak counts how many days in a row a user has completed at least one lesson, and it resets to zero the first day they miss. Publicly shared streaks have stretched into the multi-year range. Users describe protecting them on vacation, in hospitals, and on wedding days. Duolingo has also said publicly that active streak users come back at materially higher rates than non-streak users. The streak works not because the lessons are intrinsically thrilling, but because breaking a long-running record feels like losing something the user personally built.

Gacha games take the same mechanic and monetize it at industrial scale. In recent years, top titles like Honkai: Star Rail and Genshin Impact have still been generating hundreds of millions in player spending through banner systems that reward players for continuing the same chase rather than walking away. Once a player has spent $200 pursuing a single character, the rational move is to stop. The sunk-cost move is to spend $200 more, because the new cost feels smaller than the cost of “wasting” the first $200. Gacha banner design turns that decision into the only visible option on screen.

Facebook was the first consumer product to build a social-graph version of this mechanic at global scale. Every day on Facebook, users build something that would vanish if they quit: friendship connections, photos, videos, shared memories. I personally have many friends I hang out with in real life, but I do not have their phone numbers or email addresses. The only way for me to contact these people is through Facebook. You can imagine how hard it is to quit, even if I no longer feel good using it. Facebook holds your social graph hostage. The algorithm may frustrate you. But the cost of leaving is losing years of accumulated relationships that exist nowhere else.

The Google Counter-Example

Compare Facebook to Google’s search engine. Google is extraordinarily popular, but for most of its history it did not build up meaningful things you would lose by quitting. It just happened to be the best tool available. It takes one change of mind: “today I’ll search on Bing instead”: and if everyone had that thought, Google could lose traction overnight. There is no accumulated investment keeping you there.

Google’s response was personalized search results, which learn your preferences the longer you use them. Switching means losing the personalization. This is a mild Sunk Cost Prison built through accumulated data. It is not as strong as Facebook’s “we have your friends hostage,” but it is a step in the retention direction. The interesting wrinkle in 2026 is that a new generation of users is already fluent in jumping between Perplexity, ChatGPT Search, Google, and Claude for different queries. The personalization moat Google spent two decades building is weaker when users expect to consult four AI search tools in a single afternoon. Search is one of the few consumer domains where the Sunk Cost Prison is losing its walls.

The Three-Question Ethical Test

Whenever I am asked to consult on a design that uses the sunk cost fallacy, I run it through three questions before giving an opinion.

1. Would the user endorse this mechanism if they saw how it worked? Would a user who read the source code or the design doc shrug and keep using the product, or would they feel manipulated? If the mechanic only survives by being invisible, it fails the first test.

2. Does the activity the user is locked into actually serve the user? A streak that makes someone learn Spanish every day serves the user. A streak that makes a teenager send an empty snap at 2 a.m. to a friend she no longer cares about does not. The test is whether the thing being protected is truly valuable to the person being locked in, not only to the business locking them in.

3. Can the user exit cleanly? Is there a way out that does not require destroying the accumulated value? Data export, social-graph transfer, or the ability to pause rather than lose all count as clean exits. A cage with a door is ethically different from a cage without one, even if very few users use the door.

If the answer to any of these three is no, the design is a trap, not a prison. I will not help build it.

If you want the longer ethics protocol I run on consulting engagements (covering migration paths, sunset criteria, and identity-cost reframing), I cover it in the Octalysis Prime masterclass.

The Anti-Sunk-Cost Move: Designing a Clean Exit

The single best anti-sunk-cost design move is the Streak Freeze. Duolingo, the same company that built one of the most powerful Sunk Cost Prisons in consumer software, also lets users pause their streak for a day without losing it. On the surface, this weakens the mechanic. In practice, it strengthens it. A user who knows they can safely protect their streak during a sick day or a long flight does not panic-quit the app when life gets in the way. Designers can strengthen this further by adding a Hunter’s Mark (GT #140) pre-commitment step so the user declares the week ahead instead of just reacting to it. They stay. The streak survives. The relationship between user and product feels collaborative rather than adversarial.

If you are going to build a Sunk Cost Prison, build an emergency exit inside it. Users stay longer in a cage that has a door than in a cage that does not. This is not a trick: it is the part of the research literature that most designers forget.

Sunk Cost Prison vs. FOMO Punch: Same Core Drive, Different Phases

The mistake I see over and over is designers treating the Sunk Cost Prison and FOMO Punch (Game Technique #84) as interchangeable CD8 tools. They are not. They live in different phases of the user journey, and swapping them is one of the most expensive mistakes a gamification designer can make.

FOMO Punch belongs to Discovery and Onboarding. At the start of a user’s journey there is nothing accumulated yet, so the only loss-based hook available is the loss of something the user does not yet have. Limited drops, closing windows, expiring invites: these are FOMO Punch moves, and they are what get a stranger to sign up in the first place.

Sunk Cost Prison belongs to Scaffolding and Endgame. By then the user has already accumulated something real: a streak, a friend graph, a set of playlists, a character roster. Only at that point is there something to be held hostage. Running the Sunk Cost Prison on a day-one user is trying to imprison someone inside a room that has not been built yet.

A mature CD8 design uses FOMO Punch to get users in the door and the Sunk Cost Prison to keep them inside once they have built enough to lose. Reversing the order almost never works.

Escaping the Sunk Cost Prison: The FOMO Punch in Practice

Here is something I teach in my Octalysis masterclass that surprises people: the best way to escape a Sunk Cost Prison is through another Core Drive 8 mechanism, the FOMO Punch.

The Sunk Cost Prison says: “You’ll lose what you already have.” The FOMO Punch says: “You’ll lose what you could have.” When the fear of missing future opportunities exceeds the pain of abandoning past investment, people break free.

I experienced this personally with my Diablo II addiction. I had invested thousands of hours. Walking away meant “wasting” all that time. But I asked myself: “When I’m 72, will I still be playing Diablo II? If I was, I’d be a sad, sad loser. Probably not at 40 either. Maybe not at 25. So why play five more years from now? Why next year? Why now?” The FOMO Punch of missing the better part of 60 years of real life outweighed the sunk cost of virtual progress. I quit that day.

When you design your own experience, think about both sides: what makes users reluctant to let go (sunk cost), and what might eventually pull them away (FOMO Punch). If you can provide the pull within your own ecosystem: new features, new seasons, new social opportunities: you convert Black Hat retention into something closer to genuine engagement.

Practical Steps for Applying the Sunk Cost Fallacy Ethically

1. Map what the user has invested, not what you have spent. The sunk cost that matters is the user’s: their streak, their friend graph, their saved playlists, their character collection. If you cannot name it in one sentence, the mechanic is not yet load-bearing.
2. Wait for the Scaffolding phase. The Sunk Cost Prison only works once there is something to be locked inside. On Day 1 of the user journey, use FOMO Punch instead; reserve sunk-cost mechanics for users who have been with the product long enough to have built something real.
3. Pair every CD8 mechanic with at least one White Hat Core Drive. Epic Meaning (CD1), Accomplishment (CD2), or Empowerment (CD3) should supply the forward-looking reason to stay. Sunk cost alone produces resentful users who leave the moment a competitor offers a migration path.
4. Build a visible emergency exit. The Streak Freeze is the canonical move. Data export, graph portability, and pause-without-loss options all qualify. A cage with a door is ethically different from a cage without one: and empirically retains users better.
5. Run the three-question ethical test before shipping. Would the user endorse the mechanism if they understood it? Does the protected activity serve the user? Is a clean exit available? Three yeses, ship. Any no, redesign.
6. Pre-mortem on the business side. Before building a sunk-cost mechanic, write down the conditions under which you would remove it. A mechanic with pre-specified retirement criteria is far less likely to become a dark pattern over time.
7. Review for identity cost, not just financial cost. The most sticky sunk-cost mechanics are ones users have built identity around. Design for graceful identity migration when it is time for them to leave: promote the growth, not the abandonment.

The Sunk Cost Fallacy Was the Beginning, Not the End

Thaler’s 1980 paper and Arkes and Blumer’s 1985 experiment are 45 and 40 years old respectively. In those four decades, the sunk cost fallacy has gone from an economic curiosity to a foundational concept in decision research, a staple of undergraduate psychology curricula, and: in the last fifteen years: one of the most heavily deployed behavioral hooks in consumer software. Duolingo, Snapchat, every gacha publisher, every major social network, every streak-based fitness app, and every retention-optimization team at every subscription product has either named the mechanic or rediscovered it under a different label.

Research is still moving. The neuroscience is now precise enough to explain why abandoning a sunk cost activates the same brain regions as physical pain. Developmental work keeps refining the story about when the bias emerges. Cross-cultural studies continue to find cases where the fallacy varies by individualism, stress level, or time horizon. The frontier question: the one the current generation of behavioral designers has to answer: is not whether sunk cost exists. It does. The question is whether we will keep building products that depend on it to retain users, or whether we will design the sunk-cost exits our users deserve.

If you are designing a product, a policy, a service, or a team, the sunk cost fallacy is one of the most powerful retention forces you can activate: and one of the easiest to activate without meaning to. Use it deliberately, pair it with White Hat motivation, build the emergency exit, and run the three-question ethical test before shipping. Do that, and the fallacy becomes a safety net for users who would have quit during a slow week. Skip that, and the fallacy becomes a cage the user will eventually burn down on the way out.

If you want to go deeper on the architecture of Core Drive 8 and the design ethics of using sunk-cost mechanics responsibly, the longest-form treatment is in my book Actionable Gamification — and if you want to apply this on your own product with feedback from the Octalysis team, the Octalysis Prime masterclass is where I teach the Sunk Cost Prison and FOMO Punch interplay live. The free starting point if you are new here is the Core Drive 8: Loss & Avoidance primer.

Frequently Asked Questions About the Sunk Cost Fallacy

What is the sunk cost fallacy in plain English?

The sunk cost fallacy is the tendency to keep investing time, money, or effort into something because of what you have already spent, rather than because of what you can still gain. The money and time already spent are gone and cannot be recovered by any future decision, so rationally they should not affect what you do next. In practice, humans treat the past expenditure as a reason to continue: which is why people finish bad movies, stay in failing projects, and hold onto losing stocks.

Who discovered the sunk cost fallacy?

Economist Richard Thaler formally named the bias in his 1980 paper “Toward a Positive Theory of Consumer Choice”. Psychologists Hal Arkes and Catherine Blumer produced the first rigorous empirical demonstration in 1985, using a theater-ticket field experiment at Ohio University. Organizational behaviorist Barry Staw had already identified the related phenomenon of escalation of commitment in his 1976 “Knee-Deep in the Big Muddy” paper. Together these three researchers mapped the individual, organizational, and economic dimensions of the effect.

What is the difference between the sunk cost fallacy and loss aversion?

Loss aversion is the more general finding that losses feel roughly twice as painful as equivalent gains feel pleasant. The sunk cost fallacy is a specific downstream consequence: because the brain refuses to fully re-code a past expenditure as already lost, it continues to treat that expenditure as worth protecting. Loss aversion is the engine; sunk cost is one of the vehicles it drives. You can be loss-averse without committing the sunk cost fallacy, but you cannot commit the sunk cost fallacy without loss aversion underneath.

Is the sunk cost fallacy always irrational?

Not automatically. Continuing a project because of prior investment is irrational when the prior investment does not change the forward-looking expected value. It is rational when prior investment affects credibility, learning curves, switching costs, or relationship value: which it often does in the real world. The fallacy is a systematic bias, not a guarantee of a bad decision. Good behavioral design separates the cases where persistence is truly forward-looking (finishing a college degree, completing a language course) from the cases where persistence is only protecting a past expenditure.

Why do older adults commit the sunk cost fallacy less often?

JoNell Strough and colleagues have shown in multiple studies that adults over 60 are significantly less susceptible than young adults. The likely reasons are experience and a different relationship to finite time. Older adults have lived through more failed projects and have learned what sunk-cost persistence feels like in the rearview mirror. They also tend to weight remaining future time more heavily, which makes continuing a losing course feel more clearly like spending what cannot be replaced. Young adults have the opposite frame: they have future time in abundance and less experience with the cost of chasing losses.

What is escalation of commitment?

Escalation of commitment is the sunk cost fallacy operating at the organizational and social level. Introduced by Barry Staw in 1976, it describes the tendency of decision-makers to invest additional resources into a losing course of action when they personally authorized the original investment. Two amplifiers turn the individual bias into an organizational force: self-justification (admitting the first call was wrong is admitting I was wrong) and audience effects (private write-offs are cheap, public write-offs are expensive). The Concorde aircraft and the Vietnam War are the textbook real-world cases.

What is the Concorde fallacy?

The Concorde fallacy is a colloquial European term for escalation of commitment, named after the Concorde supersonic jet. The Concorde’s development budget ballooned from an original £70 million estimate in 1962 to over £1.1 billion by the jet’s 1976 launch: a more than tenfold overrun both the British and French governments knew was economically indefensible by the late 1960s but could not abandon because of the sunk political and financial investment. The term has stuck because the project is such a clean example of a capable organization unable to kill a losing line of work.

What is the counter-move against a sunk cost fallacy trap?

In the Octalysis Framework, the counter-move is FOMO Punch (Game Technique #84): reframing the decision around the loss of a future opportunity rather than the preservation of a past one. Fear of missing what comes next reliably beats fear of losing what is already spent. At the personal level, the same move works: stop asking “would leaving waste the last ten years?” and start asking “what do I lose if I stay in this for the next ten?” The second question is the one with information in it, because the second ten years are the only ones you can still shape.

How do designers ethically use the sunk cost fallacy?

By pairing it with a visible emergency exit and at least one White Hat Core Drive, and by running the three-question ethical test before shipping: (1) would users endorse the mechanism if they understood it, (2) does the protected activity actually serve the user, and (3) can the user exit cleanly without destroying the accumulated value? The Duolingo Streak Freeze is the canonical example of a sunk-cost mechanic done ethically: the streak creates real retention pressure, and the Freeze gives a user a humane way out on a bad day without losing their streak entirely.

What are the best books and papers to read about the sunk cost fallacy?

Start with Thaler’s 1980 paper “Toward a Positive Theory of Consumer Choice” and Arkes and Blumer’s 1985 paper “The Psychology of Sunk Cost” in Organizational Behavior and Human Decision Processes. Kahneman’s Thinking, Fast and Slow (2011) puts the bias in the broader context of System 1 and System 2 thinking. For the design side, my own Actionable Gamification maps sunk cost into Core Drive 8: Loss & Avoidance as Game Technique #50: the Sunk Cost Prison, and walks through the ethical test above.

References

1. Thaler, R. H. (1980). Toward a Positive Theory of Consumer Choice. Journal of Economic Behavior & Organization, 1(1), 39–60.
2. Arkes, H. R., & Blumer, C. (1985). The Psychology of Sunk Cost. Organizational Behavior and Human Decision Processes, 35(1), 124–140.
3. Staw, B. M. (1976). Knee-Deep in the Big Muddy: A Study of Escalating Commitment to a Chosen Course of Action. Organizational Behavior and Human Performance, 16(1), 27–44.
4. Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263–291.
5. Garland, H. (1990). Throwing Good Money After Bad: The Effect of Sunk Costs on the Decision to Escalate Commitment to an Ongoing Project. Journal of Applied Psychology, 75(6), 728–731.
6. Arkes, H. R., & Ayton, P. (1999). The Sunk Cost and Concorde Effects: Are Humans Less Rational Than Lower Animals? Psychological Bulletin, 125(5), 591–600.
7. Friedman, D., Pommerenke, K., Lukose, R., Milam, G., & Huberman, B. A. (2007). Searching for the Sunk Cost Fallacy. Experimental Economics, 10(1), 79–104.
8. Strough, J., Mehta, C. M., McFall, J. P., & Schuller, K. L. (2008). Are Older Adults Less Subject to the Sunk-Cost Fallacy Than Younger Adults? Psychological Science, 19(7), 650–652.
9. Brockner, J. (1992). The Escalation of Commitment to a Failing Course of Action: Toward Theoretical Progress. Academy of Management Review, 17(1), 39–61.
10. Odean, T. (1998). Are Investors Reluctant to Realize Their Losses? Journal of Finance, 53(5), 1775–1798.
11. Sweis, B. M., Abram, S. V., Schmidt, B. J., Seeland, K. D., MacDonald, A. W., Thomas, M. J., & Redish, A. D. (2018). Sensitivity to “Sunk Costs” in Mice, Rats, and Humans. Science, 361(6398), 178–181.
12. Haller, A., & Schwabe, L. (2014). Sunk Costs in the Human Brain. NeuroImage, 97, 127–133.
13. Bornstein, B. H., & Chapman, G. B. (1995). Learning Lessons from Sunk Costs. Journal of Experimental Psychology: Applied, 1(4), 251–269.
14. Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
15. Chou, Y. (2015). Actionable Gamification: Beyond Points, Badges, and Leaderboards. Octalysis Media.

Related Reading

• The Octalysis Framework: Complete Gamification Framework
• Core Drive 8: Loss & Avoidance
• FOMO Punch (Game Technique #84)
• Black Hat vs. White Hat Behavioral Design
• How the 8 Core Drives Interact

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?