We have presented the theory behind an effective method to improve calculation efficiency for heterogeneous trading books, along with a case study in its application. Our case study focussed on XVA, but the technique should be effective for any exotic book with raw calculation statistics. With low implementation cost, we can take advantage of inefficiencies in standard Monte Carlo setup for large books to achieve speed-ups comparable to those available by AAD. The theory developed here can be extended in various ways. To take an example, although we have used a Monte Carlo setup as an illustration, it could be readily extended to PDE calculations given suitable assumptions about the scaling of errors with calculation time.