Limitations of R-Squared in Model Evaluation

Imagine you're evaluating how effectively a model predicts data trends. You aim to establish a connection between two variables. What are the potential drawbacks of relying solely on the R-Squared (R²) value for this analysis?