Based on more than two years of research, our relevance-based approach to prediction reinterprets statistical regression in terms of learning from experiences (as opposed to focusing on “variables”). We extend the idea to focus on predicting from the most relevant subset of past experiences, and we show how each individual prediction has a fit or confidence associated with it that tells you much more than just a model’s average R-squared. This approach offers more transparency and flexibility into making predictions compared to traditional models, offering a compelling alternative to both linear regression and machine learning.