Text similarity techniques allow for close, but not exactly, matching strings to be compared and extracted from bodies of text. This functionality can be considered very useful in automated processing of the documents. In this paper existing algorithms such as Cosine similarity, Levenhstein distance, pre-trained models etc. are compared and summarised. Based on the attorney clauses from the banking sector—the official formats and given template—we consider the effectuality of each of them. An algorithm selection is made not only based on the similarity score, but also the simplicity of the given solution, often considered an advantage in a highly regulated industry. Nevertheless, this study demonstrated that pre-trained models, in certain instances, exhibit a performance that is twice as effective as other techniques that are more readily explicable in a business context.