Count the Steps In, Then the Steps Out
Friction is not the enemy. Asymmetric friction is, and it is measurable without any psychology at all. Count the steps in and count the steps out.
Key takeaways
- Richard Thaler named sludge in Science in 2018: friction that mucks things up and makes wise decision-making and prosocial activity harder, which he calls nudging for evil.
- Sludge works by being individually defensible and cumulatively decisive. No single step in a cancellation flow is unreasonable.
- A survey of 1,591 people found the most-resented types are outdated websites with broken links, unfriendly staff, jargon-laden documents and hard-to-navigate sites.
- People with poor mental health, a tendency to procrastinate, and a lack of time or mental energy reported more vulnerability. Sludge taxes whoever has least capacity.
- Gabaix and Laibson show why it persists: in a market with inattentive consumers, shrouding can be stable, and the firm that unshrouds may not profit from doing so.
Some cancellation flows make you telephone a human being. Others make you type a phrase out by hand. Both were documented last year in a study of news subscriptions across four countries, and neither is an accident of engineering.
So here is the whole practical instruction, before any of the evidence for it. Count the steps in. Count the steps out. Whatever the two numbers turn out to be, the gap between them needed no psychology to produce and needs none to interpret, because the same team built both flows and the difference between them was a decision somebody made.
Three things follow from that count, and you can stop reading after them. Cancel at the start of a billing period rather than the end, because retention flows are built to run out your clock. Write down what you are cancelling, and why, before you begin, because retention offers work by making you re-evaluate mid-flow, at the point where you have spent effort and are quietly looking for a reason to stop. And treat the sign-up as the moment to act, since the cheapest cancellation is the subscription you set a reminder about on the day you started it.
The rest of this is why we think the count is fair, which matters, because the defence of friction is real and not stupid. A confirmation step stops an accidental purchase. An identity check stops fraud. A cooling-off page stops the impulse buy you would have regretted, which is precisely the kind of choice architecture this publication has defended elsewhere. The case has to be that the good kind appears on the way out and rarely on the way in, and that took about twenty years to establish.
2006, the cousin that does not fit
Start with the idea everyone reaches for first, because it is the one to clear away.
Xavier Gabaix and David Laibson showed that when part of a price is hidden, a market containing inattentive consumers can settle into hiding it permanently, and the firm that competes by revealing the hidden part may be unable to profit from doing so. That is a result about prices, derived in a model.
An exit flow is not a hidden price. Anybody who attempts one experiences all of it, which is the opposite of shrouded, and the curse of debiasing turns on an add-on market that cancellation does not have. So the shrouding literature is an analogy here and not an explanation. We say so because the resemblance is close enough to be quoted as though it were more.
What survives the transfer is a shape rather than a mechanism. Practices that look like individual bad behaviour often persist because a market tolerates them, and the firm that fixes one first may not be rewarded for it. Whether that is what holds exit friction in place has not, as far as we can find, been modelled or measured. Somebody should.
2008, the argument this inverts
Richard Thaler and Cass Sunstein argued that improving the environment in which people choose lets them make wiser choices without restricting any options, and that the test of a conscientious choice architect is whether the people being nudged end up better off as judged by themselves.
Keep hold of that last clause. It is the standard the next decade failed.
2018, the name arrives
Ten years on, in a single page in Science, Thaler named the inverse. Sludge is the same machinery aimed the other way: his phrase for it is nudging for evil, friction that mucks things up and makes good decisions and prosocial activity harder.
The definition does useful work because it is not about friction in general. It is about who the friction serves. A step that protects you is not sludge. The identical step, placed only on the exit, is.
And it explains why the argument is so hard to win in public. Joining is one screen, often one click, with the payment details already stored. Leaving is a settings page that is not where settings normally live, then a retention offer, then a survey asking why, then a second retention offer, then a confirmation, then an email you have to open. Defend any one of those in isolation and you will succeed. The survey is market research. The retention offer is a discount, which is a benefit. The confirmation prevents mistakes. Sludge is individually defensible and cumulatively decisive, and an argument conducted one step at a time is an argument the designer wins.
2020, somebody finally ran the experiment
Everything above this point is definition and economics. Our own replication piece would object that a named effect with no randomised test behind it is exactly the kind of claim that has been falling over for a decade. The honest answer is that the randomised evidence exists for the acquisition half of this and not for the exit half, and the gap matters enough to set out carefully.
Jamie Luguri and Lior Strahilevitz ran two large-scale experiments on representative samples of American consumers. Users exposed to mild [dark patterns](https://thewhyweproject.com/articles/what-are-dark-patterns-in-web-design) were more than twice as likely to sign up for a dubious service as the control group. What counts as a dark pattern in the first place is set out in What are dark patterns in web design. In the aggressive condition they were almost four times as likely. Aggressive patterns produced a backlash; mild ones did not, which is the commercially important half.
Their second study sorted the techniques by how well they worked. Hidden information, trick questions and obstruction were the most effective. Loaded language and bandwagon effects worked moderately. Must-act-now messages did not work at all.
Now the limitation, which is large and which we nearly wrote past. Both experiments measured getting people to sign up. This article is about getting people out. Obstruction as they tested it is obstruction placed on the way in, so their ranking establishes that obstruction works as a manipulation, and does not establish anything about exit flows specifically. What it removes is the objection that none of this has ever been tested against a control group. What it does not do is test our actual claim, and nobody has.
One result of theirs is worth carrying anyway. Where dark patterns were used, the price of the service became immaterial to whether people bought it. Design beat price, in the one direction anybody has measured.
2021, the economists draw a line
Sina Shahab and Leonhard Lades pressed on what separates sludge from ordinary cost, and concluded the two are related but distinct concepts.
The distinction is worth carrying because it is the one that survives contact with a press office. A transaction cost is a cost of doing business. Sludge is a cost somebody chose to put in your way and could remove tomorrow.
2024, what the barriers actually are
Ashley Sheil, Güneş Acar, Hanna Schraffenberger, Raphaël Gellert and David Malone characterised the subscription and cancellation flows of popular news websites in four countries. They found flows with intentional barriers: being forced to telephone a representative, or to type a phrase out by hand. They also found subscription flows that did not adequately inform people they were signing up to a recurring charge.
That is the study behind this article's opening, and it is worth being exact about what it is. It is a characterisation of design features across a sample of sites, not a count of steps and not a measurement of how many people give up. Nobody has published the second thing, which is the number everyone actually wants.
2025, who it lands on
The most recent piece of evidence is also the one that should change how you read the rest, and it is where the argument stops being about design and starts being about people.
Lades, with Margaret Samahita, Lucie Martin, Cass Sunstein and Martin Bækgaard, surveyed a nationally representative sample of 1,591 people in Ireland about which types of sludge they felt most vulnerable to and which they found least acceptable. The two tracked each other. Most vulnerability was reported to outdated websites with broken links, unfriendly staff interactions, complex documents full of jargon and hard-to-navigate sites, and those were also the types people found least acceptable.
Less vulnerability was reported to long waiting times and to requirements to provide private information, which is interesting in itself, because waiting is the sludge people complain about most loudly and apparently not the sludge that defeats them.
Two results deserve more attention than they got. People with poor mental health reported being more vulnerable to sludge and less accepting of it. And a tendency to procrastinate, along with a lack of time and mental energy, predicted more reported vulnerability.
Read those carefully, because the instrument is a questionnaire on both sides. This is people's perception of their own vulnerability, not a measurement of who actually ends up trapped in a subscription. What it establishes is that the burden is felt unevenly, and by the people you would least want it to fall on. Whether it lands unevenly is a different study, and nobody has run it.
The same survey found only minor differences in how people perceived sludge from the public as against the private sector, which suggests the objection is to the friction itself rather than to who is imposing it.
The rule, and what it rules out
Count the steps in and count the steps out. There is no threshold in the research at which a gap becomes evidence of intent, and we are not going to invent one. What the count gives you is a question to put to a company, and a question is harder to deflect than a complaint.
It also rules something out. This is not a consumer education problem. The people who report feeling it most are the ones with the least time and mental energy, and telling them to be more vigilant asks for exactly the resource the design was built to consume.
Sources
Luguri, J., & Strahilevitz, L. J. (2021). 'Shining a light on dark patterns.' Journal of Legal Analysis, 13(1), 43-109. Two large-scale experiments on representative samples of American consumers. Mild dark patterns more than doubled the rate of signing up for a dubious service against control; aggressive ones made subscription almost four times as likely and produced a backlash that mild ones did not. Hidden information, trick questions and obstruction were the most effective techniques; must-act-now messages did not work. Where dark patterns were used, the price of the service became immaterial. doi:10.1093/jla/laaa006
Sheil, A., Acar, G., Schraffenberger, H., Gellert, R., & Malone, D. (2024). 'Staying at the roach motel: cross-country analysis of manipulative subscription and cancellation flows.' CHI Conference on Human Factors in Computing Systems. Characterises subscription and cancellation flows of popular news websites in four countries; finds cancellation flows with intentional barriers such as forcing users to call a representative or type in a phrase, and subscription flows that do not adequately inform users about recurring charges. doi:10.1145/3613904.3642881
Thaler, R. H. (2018). 'Nudge, not sludge.' Science, 361(6401), 431. Defines sludge as activities that are essentially nudging for evil, mucking things up and making wise decision-making and prosocial activity more difficult, against the Nudge standard of helping people choose better as judged by themselves. doi:10.1126/science.aau9241
Lades, L. K., Samahita, M., Martin, L., Sunstein, C. R., & Baekgaard, M. (2025). 'Vulnerability to and acceptability of different types of sludge.' Behavioural Public Policy. Nationally representative survey of 1,591 participants in Ireland; most reported vulnerability to outdated websites with broken links, unfriendly staff interactions, complex jargon-laden documents and hard-to-navigate websites, which were also the least acceptable; less vulnerability reported to long waiting times and private information requirements; only minor differences between public and private sector; poor mental health, procrastination and lack of time and mental energy predicted more vulnerability. doi:10.1017/bpp.2025.10020
Shahab, S., & Lades, L. K. (2021). 'Sludge and transaction costs.' Behavioural Public Policy. Sludge and transaction costs are related but distinct concepts. doi:10.1017/bpp.2021.12
Gabaix, X., & Laibson, D. (2006). 'Shrouded attributes, consumer myopia, and information suppression in competitive markets.' Quarterly Journal of Economics, 121(2), 505-540. Shrouding may occur in an economy with some myopic or unaware consumers; where add-ons have close substitutes a curse of debiasing arises and firms cannot profitably debias consumers by unshrouding. doi:10.1162/qjec.2006.121.2.505