Nonlinear Pricing and Revealed Salience

On a major e-commerce platform, we randomize linear promotions applying a δ% discount and nonlinear ones applying the same discount above a spending threshold of M€. We unveil a striking fact: nonlinear schemes create bunching at M but no global self-selection above M relative to linear schemes with the same discount δ. This pattern rejects optimization frictions and standard biases. We offer a salience account: upon seeing (M, δ) some people overly focus on M and cut spending above M (where the threshold should not matter), others overly focus on δ and seek discounts below M (where they do not qualify). Using browsing data on attempts to redeem promotions below M, we confirm the model's predictions and show its ability to match the spending anomalies.