The Irrational Edge, Edition 3. On why canceling a subscription is so hard, and what that friction really costs.
On December 20, 2023, the New York State Attorney General filed a petition in State Supreme Court, New York County. The case was People of the State of New York v. Sirius XM Radio Inc., Index No. 453325/2023. Letitia James was the AG. The petition ran to dozens of pages. It listed seventeen sworn consumer affidavits. One of them was a handwritten letter.
The letter described a phone call. A 92-year-old woman in New York had tried to cancel her SiriusXM subscription. Her daughter, who watched the call, wrote afterwards to describe what she saw. The mother had been on the phone for nearly forty minutes.
The petition did not reproduce the letter. It paraphrased. The press release the AG’s office issued used the word maddening.
That was one of the seventeen.
Another belonged to a man named Jacob S. Robertson. Robertson had also tried to cancel. He used the online chat. The chat went forty minutes. Robertson asked to cancel. The agent kept the chat open. He asked again. The agent worked through what SiriusXM internally called a six-part script. By the end, Robertson had answered preliminary questions about his listening habits, sat through an exercise designed to make him reconsider what he was about to give up, fielded an offer to change his subscription term, fielded an offer to switch to a different full-priced plan, and fielded an offer of a discount.
The chat ended. Robertson thought he had cancelled.
SiriusXM kept charging him.
When he filed a complaint, the company told him it couldn’t locate any cancellation request from him.
By the next morning, the case was being covered by Business Insider, CBS, and Variety. SiriusXM issued a statement that afternoon. The long wait times the AG cited, the company said, were anomalies from the COVID-19 pandemic. By 2021, online chats were being answered within 36 seconds to 2.4 minutes. The numbers were narrower than the AG’s. The argument was that the system worked except when it didn’t.
Why companies make canceling so hard
Here is a question that doesn’t have an obvious answer.
SiriusXM had thirty-four million subscribers when the AG filed her case. Two million of them were in New York. The company’s annual reports described retention rates that any subscription business would envy. By the standard metrics of the industry, SiriusXM was a healthy business with loyal customers.
How do you reconcile that picture with this one? Seventeen affidavits, hundreds of consumer complaints to the AG, hundreds more to the Better Business Bureau, a six-part script the company’s own training documents called the path to keeping a subscriber on a full price plan.
The retention numbers were real. So were the affidavits. They described two different things and they were both true.
The subscription economy runs on retention. Net revenue retention, monthly churn, lifetime value. Investors price subscription companies primarily on those numbers because the math is straightforward. A customer who stays is worth more than a customer who leaves. A company that keeps its customers is worth more than a company that doesn’t.
Implicit in that valuation is a model of why customers stay. They stay because the product is good. They stay because they value it. They stay because alternatives are worse, or because switching is costly, or because the relationship has compounded into something useful. These are all real reasons. They explain a lot of what a high retention number means.
In 2025, three economists, Liran Einav, Ben Klopack, and Neale Mahoney, published a paper in the American Economic Review called “Selling Subscriptions.” They had access to data from a large payment card network covering millions of accounts. The data let them do something the rest of us can’t do. They could compare cancellation behaviour in months when customers’ credit or debit cards were replaced, when expiration or loss forced an active renewal decision, against months when the renewal was passive.
The numbers weren’t subtle. In the inattention model the authors used, consumer inertia raised seller revenues by an average of 87 percent. The range across different subscriptions ran from 14 percent to over 200 percent. The paper’s abstract put the finding plainly. Cancellation friction roughly doubles seller revenues on average.
Read that again with the SiriusXM petition in mind. If your retention rate is high, somewhere between half and most of that number is structural inertia. The cost of leaving is doing the work.
Nudge, sludge, and the Roach Motel
In August 2018, Richard Thaler published a one-page article in Science. Thaler had won the Nobel Prize in Economics the previous year for his work on behavioural economics. He had spent a career arguing that small changes in how choices are presented can shift behaviour without changing what people want. He called these small changes nudges.
The 2018 paper introduced the opposite term. He called it sludge. The article wasn’t subtle. Its title was “Nudge, Not Sludge.”
A nudge removes friction from things people want to do. Sludge adds friction to things people want to do but companies or governments would rather they didn’t. The definition Thaler offered later had two parts. First, friction. Second, bad intentions. A government form that takes ten pages to fill out is friction. A government form that takes ten pages because the agency is hoping you give up is sludge.
By that definition, Robertson’s chat was sludge. The training document that instructed agents to “think of every ‘No’ simply as a request for more information” was sludge. The instruction to build value of an existing subscription before allowing the customer to leave was sludge. The five rounds of retention offers, the order they appeared in, the rule that an agent should never skip an offer. All of it. The friction was deliberate. The intent was visible to anyone who read the script.
Once you have the word, you start to see the thing. And once you start to see the thing, you start to notice that nobody in the subscription industry talks about it. The trade press talks about retention. About customer success. About reducing churn. The phrase cancellation friction barely appears in the marketing literature. The mechanism that may account for half of subscription industry revenue doesn’t have a name marketers use.
Thaler’s paper had a closing line. Eight words. Less sludge will make the world a better place. He was writing about government forms. He had no idea, in 2018, that the sentence would land harder over the following seven years than he could’ve known.
The clearest documented case of sludge as design choice is Amazon. In June 2023, the Federal Trade Commission filed suit against Amazon in federal court in Seattle. The complaint described the cancellation flow for Amazon Prime as a four-page, six-click, fifteen-option process. Inside Amazon, the team that built the flow had given it a name. They called it Iliad. The reference was Homer’s poem about a war that lasted ten years.
Internal Amazon documents, which surfaced in court filings and were leaked to Business Insider, contained other words. Employees referred to the practice of unintended Prime enrollments as an unspoken cancer. One employee wrote that subscription driving was a bit of a shady world. A 2017 internal analysis showed the redesign of the cancellation flow had reduced cancellations by fourteen percent because customers gave up before reaching the final step. Amazon had made the same flow much simpler in the European Union the year before, after a consumer advocacy complaint there. American users got the simpler version only in April 2023, after the FTC investigation began.
The case went to trial. Three days in, on September 25, 2025, Amazon settled for $2.5 billion. It was the largest consumer protection settlement in FTC history.
Amazon was the largest case. It wasn’t unusual.
In May 2024, five researchers from European universities published a paper at CHI, the major academic conference on human-computer interaction. The paper was called “Staying at the Roach Motel.” The methodology was simple. The researchers built five subscriber personas, two American and three European. They used the personas to subscribe to ten major news websites in each country, then to cancel. They counted clicks.
For the New York Times, the subscription took 4 clicks. The cancellation took 7. For Bloomberg, subscription was 3 clicks. Cancellation was 10. For the Wall Street Journal, subscription was 5 clicks. Cancellation, from a Texas address, required a phone call. Across the American sample, 77.8 percent of news sites offered a special retention discount during cancellation. 82.7 percent required users to fill out a mandatory exit survey before completing it. The pattern held across countries. American sites were generally more aggressive than European ones, but the asymmetry was universal.
The asymmetry has a name in academic dark patterns research. It is the Roach Motel pattern. Harry Brignull, a British UX researcher, coined the broader term dark patterns in 2010. The Roach Motel was one of the first categories he documented. Easy to enter, hard to leave.
Click-to-Cancel and why the rule got struck down
By the autumn of 2024, the Federal Trade Commission had decided that the Roach Motel was widespread enough to outlaw. On October 16, 2024, the agency voted three-to-two along party lines to adopt a final amended Negative Option Rule. The chair, Lina Khan, called the experience consumers were having a doom loop. In an op-ed in The Hill the following week, she described the standard cancellation as hours on hold, robot scripts before reaching a person, transfers between departments, calls dropped right before they connected. “Too many Americans feel like they are stuck in an endless doom loop,” she wrote. The rule said cancellation had to be at least as easy as signup, in the same medium. If you signed up online, you had to be able to cancel online. If you signed up by app, by app.
The rule had its critics. Industry trade groups challenged it within a week of its passage. The legal challenges were consolidated in the Eighth Circuit Court of Appeals. The court heard oral argument on June 10, 2025. The rule was scheduled to take effect on July 14.
On July 8, six days before the effective date, the court issued a per curiam decision. The panel was three judges. James Burton Loken, a 1990 appointee. Ralph Robert Erickson, confirmed in 2017. Jonathan Allen Kobes, confirmed in 2018 by a tied 50 to 50 Senate vote with Vice President Pence casting the tie-breaker.
The decision vacated the rule in its entirety. The reason was procedural. The FTC had initially estimated the rule’s economic impact at less than $100 million. An administrative law judge later found that the impact would exceed $100 million. The agency accepted that finding but did not file the additional regulatory analysis the law required when the threshold was crossed. The Eighth Circuit found the procedural failure fatal.
The judges added a sentence to the opinion. They acknowledged that they were not endorsing deceptive practices in subscription marketing. Then they wrote that the procedural deficiencies of the Commission’s rulemaking process were fatal here.
They knew. They struck the rule down anyway.
The FTC reopened the rulemaking process in March 2026. As of this writing, federal click-to-cancel doesn’t exist.
The industry has a real response. It came in two voices.
The first was Commissioner Melissa Holyoak, the Republican commissioner who voted against the click-to-cancel rule in October 2024. Holyoak’s dissent argued that the rule was too broad, that some friction in cancellation flows reflected genuine consumer interest. A customer who would otherwise cancel because of one frustrating month might prefer a temporary discount to full cancellation. The five retention offers SiriusXM walked Robertson through were, on their face, available choices. The friction that surfaced those choices was the mechanism that made them visible at all. Sludge done crudely is exploitation. Sludge done well is what customer success departments call saving the customer. Drawing the line between them takes judgement.
The second voice was Justice Lyle E. Frank of New York State Supreme Court. On November 21, 2024, Frank ruled on the AG’s case against SiriusXM. He found for SiriusXM on most of the charges. The fraud claims and the deceptive practices claims were dismissed. Frank’s reasoning, in plain language: “It may be frustrating, but it is not deceptive.” SiriusXM had disclosed its cancellation procedure. It had not lied about it. It had merely made it hard. That, Frank wrote, was aggravating at most.
The AG won on one count out of five. The count was a procedural one. SiriusXM’s cancellation method, Frank wrote, was clearly not as easy to use as the initiation method. That was a violation of a federal procedural statute, ROSCA, that says cancellation should be as easy as signup.
The same procedural standard, in slightly different language, was what the Eighth Circuit struck down seven months later.
Holyoak and Frank weren’t wrong. Some friction is benign. Some retention offers are valuable to consumers. The Einav paper provides the empirical floor on this argument. When you isolate the cancellation friction effect, it doubles seller revenues on average. The argument from Holyoak and Frank can explain part of that. It can’t explain a doubling. The number is too large to be a side effect of customer-protective design choices.
Retention vs. reality: customer love or structural inertia?
Two implications fall out of this for anyone building a subscription business or auditing one.
The first is about diagnosis. Your exit asymmetry is a confession. If your signup is two clicks and your cancellation is six, the architecture is telling you something about how much you trust your product to be missed. The asymmetry says: we don’t think they’d stay if leaving were as easy as joining. Whether you decided that consciously or whether your retention team built it under pressure to hit a churn number, the result is the same. The asymmetry is on your servers. Anyone with a developer console can find it.
The second is about reading retention numbers. Your retention number is two ingredients mixed together. Customer love and structural inertia. They aren’t the same thing. They have different shelf lives. Inertia evaporates the moment a regulator forces an active choice, or a credit card replacement triggers an active renewal, or a journalist writes a story that changes how customers think about the product, or a competitor offers them an exit ramp with one click. The companies that survive the regulatory cycle are the companies whose retention is mostly love. The companies whose retention is mostly inertia find out which it was the moment something forces a choice.
The petition AG James filed on December 20, 2023 is still open. Frank’s ruling is on appeal. The federal click-to-cancel rule is back in committee, and when it returns the Eighth Circuit will see it again.
None of that changes what happens on the balance sheet right now.
One billion paid subscriptions are active in the United States. Some of them belong to people who love what they’re paying for. Some of them belong to people who once spent forty minutes on a chat or a phone call trying to leave, and could not.
Both kinds count the same. Same revenue line, same retention rate, same valuation multiple. The financial statements don’t distinguish.
More from The Irrational Edge
- The Default Effect: Why We Don’t Really Choose: how defaults quietly decide outcomes.
- Why We Idolize Then Destroy: Indonesia’s Idol Cycle: the psychology behind sudden reversals.
Leave a Reply