A parsimonious extension of a well-known portfolio credit-risk model allows us to study a salient stylized fact - abrupt switches between high- and low-loss phases - from a risk-management perspective. As uncertainty about phase switches increases, expected losses decouple from unexpected losses, which reflect a high percentile of the loss distribution. Banks that ignore this decoupling have shortfalls of loss-absorbing resources, which is more detrimental if the portfolio is more diversified within a phase. Likewise, the risk-management benefits of improving phase-switch forecasts increase with diversification. The analysis of these findings leads us to an empirical method for comparing the degree of within-phase default clustering across portfolios.