The authors consider the case for augmenting risk models to be used in portfolio construction to reflect information embedded in the portfolio manager’s alphas. They consider both smart beta models and cases in which alpha signals are partly factor driven but incorrectly perceived to be stock specific. In smart beta cases, the authors argue that mechanically augmenting the risk model can cause losses by distorting an otherwise-correct factor structure. The authors show that for cases in which asset-specific alpha signals might unexpectedly be related to hidden systematic factors, errors of omission—missing these hidden factors—generally result in larger expected losses in portfolio efficiency than do errors ofcommission—unintentionally including nonexistent “phantom” factors. When the alpha signals are very noisy, the practice of mechanically augmenting the risk model with a custom risk factor to offset that noise can improve portfolio efficiency. However, in those cases, the custom risk factor has nothing to do with underlying sources of true risk that all investors face, but instead serves as a penalty that in a back-door way tends to adjust for weak quality of the manager’s alphas.