Decision Fatigue Is Real and Your Product Is Probably Making It Worse

You’re looking at the A/B test results at 4:47 on a Thursday. The variant with the extra upsell step — the one that makes people choose a plan, then choose an add-on, then confirm they don’t want the add-on — is winning by four points on completed checkouts. You already know you’re going to ship it before you finish the thought that starts with “but this feels like a lot.” Decision fatigue in product design usually doesn’t announce itself. It shows up as a chart that’s winning.

That’s the moment worth stopping on, because the four points are real and so is what produced them. Both things can be true. This is where the research actually leads.

What Is Decision Fatigue, Exactly?

Decision fatigue is the documented decline in decision quality after a stretch of decision-making, first shown clearly outside a lab in a study of Israeli parole judges. It’s not tiredness in the colloquial sense — it’s a measurable shift toward the default, low-effort option.

In 2011, PNAS published Shai Danziger, Jonathan Levav, and Liora Avnaim-Pesso’s analysis of eight Israeli parole judges over ten months. The judges heard cases in random order, spending about six minutes on each. Right after a meal break, they granted parole roughly 65% of the time. Over the next two hours, that approval rate fell steadily — toward zero, right before the next break. Then it reset. The judges weren’t getting harsher opinions on the merits as the day went on. They were defaulting to the status quo — denying parole, the option that requires no further evaluation — as their mental resources ran down (PNAS, 2011).

That’s the actual mechanism, and it matters that it’s specific: fatigue doesn’t make people choose badly at random. It makes them choose the path that requires the least additional evaluation. In a courtroom, that’s denial. In a product, that’s whatever you’ve set as the default.

Is the Paradox of Choice Real?

Partly. The effect is real but conditional — meta-analysis finds no reliable average effect of “more options” across all situations, and it shows up dependably only when the options are hard to tell apart, the stakes feel meaningful, and the person choosing isn’t an expert in the category.

That qualification is worth taking seriously before going further, because the popular version overshot. The famous 2000 jam study, where a supermarket display of 24 jams drew more browsers but a display of 6 jams sold ten times more, doesn’t hold up as a universal law. A 2010 meta-analysis by Benjamin Scheibehenne, Rainer Greifeneder, and Peter Todd pooled 50 experiments testing the effect and found the average impact of “too many options” on satisfaction and purchase was close to zero. If you’ve seen someone online debunk the paradox of choice with this stat, they’re citing something real.

But the debunking usually stops one sentence too early. A more recent review in the psychology literature breaks down when the effect does and doesn’t appear: it emerges reliably under exactly those three conditions — similar options, real stakes, non-expert chooser (PMC, 2024). That’s not an edge case for software. That’s the exact shape of a pricing page, a settings menu, or an insurance-plan comparison inside an app — options that look similar, cost real money, and are being evaluated by someone who has no domain expertise in what they’re buying. The “it’s a myth” rebuttal is technically accurate about supermarket jam and quietly wrong about the conditions most digital products actually create.

What Are Dark Patterns in Subscription Cancellation?

Dark patterns in subscription cancellation are interface choices that make leaving harder than joining: buried cancel links, multi-page confirmation loops, pre-checked renewal boxes, retention offers wedged between the user and the exit, and confirm-shaming copy. The FTC’s term for the deliberate version is “sludge,” and its own audit found the majority of subscription products use at least one.

The 2022 FTC staff report, “Bringing Dark Patterns to Light,” catalogued the mechanics and named the intent — friction added because it causes people to become fatigued and give up (FTC, 2022). Two years later, a follow-up review of 642 subscription websites and apps found that 76% used at least one dark pattern, and 67% used more than one. Seventy percent didn’t tell users how to cancel. Sixty-seven percent didn’t disclose the date by which a user needed to act to avoid another charge (TechCrunch, 2024).

Read that next to the parole study and the pattern is uncomfortable. The mechanism isn’t guessed at — it’s the same one. A depleted decision-maker defaults to the path of least resistance. In a subscription flow, the path of least resistance is staying subscribed. The friction isn’t a bug in the cancellation page. It’s the whole design brief.

If the Research Is This Clear, Why Does Every Product Still Do This?

Because the team that ships the friction gets credit for the number that moved, and no team owns the number that got worse. That’s not a personality flaw in any individual designer or PM. It’s the incentive structure they’re operating inside.

Growth work is organized around short-window, per-team metrics: activation lift this sprint, checkout conversion this quarter, retention inside a 14-day A/B test window. Each of those metrics can go up when you add a choice, a step, or a nudge — a pre-selected add-on, a fourth pricing tier, a “recommended” badge that reframes a decision as already-made. The team that ships it sees its number improve and moves to the next sprint. What that same change costs in trust, in the mental tax paid by every user who has to evaluate one more screen, in the slow erosion that eventually shows up as churn three quarters later — none of that is attributed back to the choice that caused it. It lands in someone else’s dashboard, or in no dashboard at all.

You don’t have to infer any of this from the outside. In June 2023 the FTC sued Amazon over Prime, and the complaint made the company’s internal vocabulary public: the Prime cancellation flow was named “Iliad,” after Homer’s epic about a war that dragged on for ten years. Joining could happen in a single click during checkout — sometimes, the FTC alleged, without the shopper registering that a recurring subscription was attached to the button they pressed. Leaving went through the Iliad: multiple pages, repeated confirmations, and a sequence of save-offers inserted between the user and the thing they came to do. The agency’s characterization was blunt — the primary purpose of the flow was not to let subscribers cancel but to stop them (FTC, 2023). Someone at Amazon owned the subscriber number. Nobody owned the other one.

In September 2025, on the second day of trial, Amazon settled: a $1 billion civil penalty — the largest ever for an FTC rule violation — plus $1.5 billion in refunds to consumers (FTC, 2025). Two and a half billion dollars is what the bill looks like when it finally arrives for a long series of locally rational sprint decisions, at a company with more UX research capacity than anyone reading this.

And the cost isn’t only regulatory. Baymard Institute, synthesizing roughly 50 cart-abandonment studies, puts the average documented abandonment rate at 70.19%, and finds that 22% of US online shoppers abandoned a cart in the past quarter specifically because the checkout was too long or too complicated (Baymard, 2026). The extra step that wins your test is drawn from the same family of steps that one in five shoppers name as their reason for leaving. Your test window just isn’t long enough or wide enough to see both sides of that trade.

Call it what it is: a growth-metrics economy that rewards whoever ships the next increment of friction and never bills them for the cognitive cost. Every individual decision inside it can be locally rational — the four points are real — while the aggregate product gets measurably harder to use. Nobody in the loop is lying about their numbers. The system just isn’t set up to notice what it can’t attribute.

Is Removing the Friction Actually Worth the Revenue Risk?

Usually yes, and the downside is no longer only theoretical — it’s regulatory, even though the regulatory picture is messier than the headlines suggest.

The FTC’s click-to-cancel rule was finalized in 2024 and then vacated in full by the Eighth Circuit on July 8, 2025, days before its compliance deadline, on procedural grounds rather than substantive ones (Cooley, 2025). The rule died. The exposure didn’t. The statute underneath it, ROSCA, was never touched, and ROSCA is what the $2.5 billion Amazon order was won under. So the four points you win in a checkout test can still turn into a compliance liability a year later — the mechanism just runs through enforcement rather than a rulebook.

That’s the argument that lands in a roadmap review. But it’s worth sitting with the other one too, the one that doesn’t show up in a quarterly deck: every additional choice you ask a fatigued person to make is a small transfer of cost from your team’s sprint to their afternoon. The jam study taught people to dismiss that as folk psychology. The parole study, and the FTC’s own audit of what companies do with that finding, suggest the dismissal came too early.

You don’t need to remove every choice. You need to ask, for each one, whether it exists because it serves the person making it or because it moves a number nobody’s asked to defend past this sprint.

You’ve got a minute before the standup where you say whether the upsell step ships. Here’s the whole note: the four points aren’t fake, but they’re not free — someone paid for them in exactly the currency the parole judges were running out of by 3 p.m. Ship it if you can look at the cancellation flow next to it and still call the whole thing fair. If you can’t, that’s your answer, and you already know it’s your answer. Say it in the standup.