Door-in-the-face

The original demonstration pairs a request nobody takes with one that sounds reasonable. Cialdini, Vincent, Lewis, Catalan, Wheeler and Darby (1975) asked students to counsel juvenile delinquents two hours a week for two years and, once refused, asked the same students to chaperone a group of those delinquents on a single zoo trip; the zoo request was accepted far more often after the refusal than when it was the only thing asked. The authors read that as reciprocal concessions: the requester gives ground, so the target gives ground back. Genschow and colleagues (2021) ran a direct replication in another country 45 years later, with the refusal-first condition beating both a target-request-only control and a small-first-request control at compliance rates similar to the original. For a compliance classic that is an unusually clean replication record, and it is the strongest thing this entry has.

Pooled, the effect is small and heavily conditional. Dillard, Hunter and Burgoon (1984) meta-analysed the sequential-request literature and put door-in-the-face at r = .15 and foot-in-the-door at r = .17, small even under optimal conditions, with two hard constraints: both worked only where the request was prosocial, and door-in-the-face worked only where the delay between the two requests was brief. They also concluded that reciprocal concessions — the mechanism the technique is named after — was seriously flawed as an account of the data, which is worth remembering before designing a flow around the idea that the user feels you have conceded something.

The split that matters for a product is between saying yes and doing the thing. Feeley, Anker and Aloe (2012) found door-in-the-face significant for verbal compliance (k = 78, r = .126) and not significant for behavioural compliance (k = 39, r = .052). Feeley's (2025) 50-year update puts verbal compliance at r = .14 across 89 comparisons and behavioural compliance at r = .08 across 53, with behavioural effects growing over the period, and reports the technique as more effective on student targets, where compliance in the control condition was low, and — for behaviour — where the request was for money rather than for volunteering. An agreement extracted in the moment is the part that replicates; the follow-through is the part that does not, which is precisely the gap between a user accepting a pause on the cancellation screen and that user still being worth anything a quarter later.

None of that record is about a company holding on to a subscription. The requests are charitable, the requester is a person in front of you, the two asks come seconds apart and share a beneficiary, and Feeley's moderator — the effect shrinks as control-condition compliance rises — points the wrong way in a cancel flow, where the person has already decided before the screen loads. There is no published controlled experiment on pause-before-cancel or on a downgrade offered at the exit, and the save rates quoted by subscription-platform vendors have no control group and no holdout. One fair alternative is cheap to offer, and its value has to be established by your own test.

What has been measured is how the pattern ships. Sheil, Acar, Schraffenberger, Gellert and Malone (2024) completed 67 subscriptions across 34 news sites in the US, UK, Netherlands and Germany, and found special offers or discounts to stay on 51.9% of the European sites and 77.8% of the US sites, with exit surveys on about 67.1% of European sites of which roughly 72.5% were mandatory, and on nearly every US site with 82.7% mandatory. Cancelling cost more clicks than subscribing everywhere: 3.5 against 4.9 for European sites accessed from Europe, 3.2 against 6.2 for US sites, 3.5 against 5.8 for US personas on US sites, and ten clicks for one chatbot-based flow. Luguri and Strahilevitz (2021) measured what that extra friction buys: 11.3% of their control group (73 of 644) accepted an identity-theft plan, against 25.8% under mild dark patterns (155 of 600) and 41.9% under aggressive ones (217 of 518) — which included a trick question and three forced information screens, not a concession. Anger in free-text responses ran at 5.70% in the control group, 6.09% in the mild condition and 12.82% in the aggressive one. Most of the lift available in a cancel flow is obstruction, and the aggressive version is bought with ill will.

The counter-offer can also push the user out. Ascarza, Iyengar and Schleicher (2016) ran a field experiment in which a mobile operator proactively recommended a cost-minimising plan: 6% of the control group churned in the following three months against 10% of the treatment group, which they attribute to lowered inertia to switch and to making past usage salient. Putting a cheaper plan in front of someone makes the whole question of whether to keep paying live again. That is the risk a downgrade offer carries, and it is invisible to a dashboard that counts saves on the cancellation screen.

The legal position moved recently and in both directions. The FTC's 2024 Negative Option Rule, which would have required cancellation to be at least as easy as the mechanism that obtained the consumer's consent, was vacated in full by the Eighth Circuit in Custom Communications, Inc. v. FTC in July 2025 and is not in force; the Commission then recodified the narrow pre-2024 prenotification rule, effective 12 February 2026. It has not retreated on the substance: its advance notice of proposed rulemaking of March 2026 asks directly whether it is unfair or deceptive to offer discounts or other incentives to stay enrolled — the FTC calls them Saves — instead of promptly honouring a cancellation request, and how much longer cancellation takes after a consumer declines one. ROSCA and section 5 still apply in the meantime, and the Prime case shows what that is worth: Amazon settled in September 2025 for $2.5 billion, a $1 billion civil penalty and up to $1.5 billion in redress, of which more than $845 million had been paid out by September 2026. In Germany, § 312k BGB requires a “Verträge hier kündigen” button that leads the consumer directly to a confirmation page carrying a “jetzt kündigen” button, so a save-offer interstitial between the two does not meet that standard — and only 37.5% of the German sites Sheil and colleagues tested cleared the two-step bar. Across the EU, Article 9(d) of the Unfair Commercial Practices Directive counts onerous or disproportionate non-contractual barriers to exercising a contractual right, termination included, towards harassment, coercion or undue influence, and no retention screen may gate the 14-day withdrawal right in Article 9 of the Consumer Rights Directive.