Nudge Examples That Worked

Transitioned from an 'Opt-In' policy (requiring active sign-up) to an 'Opt-Out' presumed consent policy (where citizens are registered as donors unless they explicitly request otherwise).

Donor Consent Registration Rate: 18% → 99.1% (451% relative) · n = Cross-national European Comparison

Allowed forecasters to slightly edit the algorithm's predictions (up to +/- 10% or +/- 5%) before submission, rather than forcing them to accept the algorithm's recommendation as-is.

Adoption Rate of the Algorithmic Forecast: 15% → 88% (487% relative) · n = 3 experimental studies, 1,200 trials

The UK Behavioral Insights Team added a single social norm line to payment letters: '9 out of 10 people in the UK pay their taxes on time.' To make it even more salient, they localized the norm: '9 out of 10 people in your local area pay on time.'

Payment Rate within 23 Days: 67.5% → 83% (23% relative) · n = 140,000 taxpayers

Simplified the layout to 3 primary options (Credit Card, PayPal, and Express Wallet) while hiding the others behind a clean, secondary 'More Payment Methods' disclosure link.

Completed Purchase Checkout Rate: 71.2% → 84.8% (19% relative) · n = 45,000 sessions

Implemented the 'Save More Tomorrow' (SLaT) protocol: employees commit to allocating a portion of their *future* salary raises to their retirement fund, bypassing current pocket pain.

Average Retirement Saving Rate: 3.5% → 13.6% (289% relative) · n = 3 companies, 1,200 employees

Opower printed a peer-comparison chart on monthly utility bills, grading energy usage against efficient neighbors and rewarding efficiency with simple emoticons (smiley faces).

Percentage Household Energy Savings: 0.1% → 2.1% (2000% relative) · n = 600,000 households

Radiologists were given access to an AI pathology detector displaying exact probability scores (0-100%) for critical conditions like pneumonia, but were otherwise left to make final determinations.

Diagnostic Classification Accuracy: 82.5% → 79.8% (-3% relative) · n = 200 radiologists, 10,000 cases

Evaluated a machine learning model designed to predict defendant flight risk and crime probabilities, compared against the historical release rates and subsequent rearrests under human judges.

Rearrest Rate of Released Defendants: 18.2% → 13.7% (-25% relative) · n = 750,000 cases in New York City