It is an undeniable fact of the times we live in that machine learning is growing increasingly ubiquitous in a business setting. More and more frequently, tasks that were previously reserved for human labour are now entering the realm of the automatable. Despite the ballooning number of machine learning models being developed, deployed, and iterated upon, many challenges remain in assessing their performance. High variance can make it difficult to discern underlying patterns in model performance. Labelled data can be hard to come by, and labelled data spanning an entire problem domain even more difficult. But despite all this, we want to continue to improve our models, and be confident in their improvement. This project is dedicated to the exploration and comparison of potential solutions to these challenges, as applied to the particular domain of financing write-off prediction.