Many Digital Humanities studies rely on a limited number of established, widely used computational tools, such as Word Embeddings, Topic Modelling or Sentiment Analysis. While the statistical performance of these methods is reliably evaluated from computational perspective, it is often overlooked how the methods fit with the research goals in humanities or social sciences. This can lead to a difficulty in interpreting and evaluating the results. Antti Kanner discusses and shows examples of finding a use for novel computational methods within a research workflow to overcome the limitations of generic computational tools.