Foodnoms: more users started a trial or plan (+58%), but a smaller share of trials converted to paid (−15.9%). Both relative to control over two weeks in July 2024, company-reported, and bundled with other paywall changes.

A 125-point bonus (about $0.09) for the planned gym visit after a missed one added 0.40 weekly visits over four weeks — 27% above the placebo group's 1.48. Milkman et al. megastudy, 24 Hour Fitness members.

Two randomised experiments: 15.44% vs 14.63% subscribed with 7 vs 30 days (software firm, 101,024 held-out users); 0.445% vs 0.368% converted over two years with 7 vs 3 days (image editor, 680,588 users). Different businesses — not a tested optimum.

Median share of installs paying within 35 days: hard-paywall apps vs freemium apps in RevenueCat's 2025 data. Different apps that chose their own model — not a test of switching.

Mojo's share of yearly subscriptions after showing only the yearly plan, with monthly behind “View all plans”. Company-reported; baseline not given. RevenueCat retells it as a “15–20% lift”, which is a different claim.

Daily steps, between-arm differences in three small trials against a fixed 10,000-step goal or no treatment (Cal Fitness, Walkadoo, Adams 2013). In Cal Fitness both arms fell; adaptive fell less.

FitnessAI moved the paywall before onboarding and added a product video in the same change; 50% more people saw it. A founder's before-and-after account, no control group.

Subscriptions reported by Mobills after changing only its button from “Subscribe now” to “Become premium” (in Portuguese). Company-reported with the basis unstated: no comparator, window or sample given.

Mojo raised yearly prices for new users 50% in countries such as the US and Germany and reports “minimal adverse effects” on free-to-paid conversion. Company-reported, with no conversion or revenue figures.

Duolingo's family-plan bookings grew “over 100%” in 2023 over 2022 (a lower bound), and family subscribers reached about 20% of paid subscribers by Q2 2024. Company-reported.

Pinterest's Google signups when a gender question sat between Google sign-in and registration, even as activation rose about 7%. Moved into onboarding with a stated reason, completion rose 11%. Company-reported; relative or absolute not stated.

Placement tests taken by new English learners in Japan after Duolingo reworded “take a placement test” to “check your English level” (both options' copy changed). Company-reported, no base, period or retention figure.

Flu-shot uptake at a workplace clinic when the reminder mailer asked employees to write the date and time they planned to go (regression-adjusted; 37.1% vs 33.1% raw). 3,272 employees at one firm, fall 2009 (Milkman et al.).

Overall learning time after Duolingo launched leagues, as reported by its former head of product; Duolingo's S-1 says “almost 20%”. Company-reported with the basis unstated, and not tied to a test.

Duolingo's bandit against random rotation over a two-week online test; new users' next-day retention rose 2.2%. Group sizes not reported (Yancey and Settles, 2020).

Steps in the 30 minutes after a walking suggestion versus none, at moments screened as available (107% on day one, undetectable by about day 29). 37 sedentary adults over six weeks (HeartSteps, Klasnja et al.).

The only randomised test of a glanceable status display: a 2008 phone-background prototype (UbiFit Garden) over three months. Duolingo and Gratitude report better retention for widget users, with no numbers that separate the widget from who adds it.

On an exam 28 days after the semester, over massed review (179 eighth graders, one Spanish course; points or relative not stated). At Duolingo, replacing the old Leitner system cut next-day practice return 7.3%.

Body weight change at six months with daily weighing, a trend graph, weekly tailored emails and lessons, against a delayed-intervention control (91 adults, 2011; WEIGH trial). A research bundle — not a weight-loss claim for an app.

App users who ever committed to a negotiated window; 32 of those 38 stayed in bed longer than agreed on most nights. The whole Sleepcare app still beat a waitlist on insomnia severity (Horsch et al., 2015 trial).

Controlled replications that found a depression benefit beyond placebo or control (Mongrain 2012, Woodworth 2017, Gold 2023). The 2005 original reported six-month effects among the 71% who finished.

Extra net deposits a month for 122 Gimme5 beta testers given goal-setting 50 days before everyone else (intent-to-treat, August–September 2017). Only 28% of goals were reached by their deadline.

Share of Bolivian savers who made every monthly deposit when the SMS named their goal and the free insurance they kept by doing so, against no reminder (9,376 clients, 2008; average compliance 56.4%).

401(k) enrollment among non-participants when the letter added the share of same-age coworkers already enrolled, against the same letter without it (one US firm, 2008). The drop was concentrated among lower-income staff.

Loss-framed weekly points game with self-chosen goals, against a text-feedback control during weeks 9 to 16; +1,391 after the game was switched off. 500 adults at elevated cardiovascular risk (ENGAGE trial, Patel et al.).

Petit BamBou's cost per acquisition with ad variations on keyword-matched pages versus the default ad. Apple-published success story; comparison design, dates and sample not given, and part of a wider campaign restructure.