Does a More Complex ML Model Actually Give Better Results?

I compared Logistic Regression and XGBoost on the same churn dataset. The simpler model performed better, even across 5-fold cross-validation.