AI in Advertising and Marketing Practice Test 2026 - Free AI Marketing Exam Questions and Study Guide

Log loss measures how good the model’s predicted probabilities are compared with what actually happened. In binary outcomes, it uses the predicted probability of the positive class and the true outcome to compute a penalty: if the event occurs, a high probability yields a small loss, while a low probability yields a large loss; if the event does not occur, a high probability for the event yields a large loss. This makes log loss sensitive to calibration—the degree to which the predicted probabilities reflect real frequencies—and to the confidence of the predictions.

This is different from simply counting correct classifications, which would ignore how certain the model was about its predictions. It’s also not a ranking metric, which assesses ordering of predictions, nor a measure of model complexity. Lower log loss means the probability estimates are closer to the true outcomes, which is crucial when you rely on probability scores to drive decisions like targeting, forecasting, or optimization in marketing AI.