A primary focus of current enterprise work is to automate human tasks for improved efficiency. IBM has proposed a framework called “SNAP” that utilizes large language models (LLMs) to generate predictions of future business processes and predict the next action to be taken. SNAP improves the prediction performance of business process management (BPM) datasets. In a paper published on the arXiv pre-print server, IBM researchers discuss the framework and its use of LLMs to create “semantic stories for the next activity prediction”. The authors note that older AI programs are limited in their ability to predict outcomes due to their inability to capture detailed business process data, while LLMs, such as GPT-3, can transform business process attributes into coherent narratives in natural language. SNAP involves three steps: creating a story template, filling in the template with details from a business process, and training the LLM to predict the next event based on the completed stories. In their research, Oved and team discovered that semantic stories significantly improve the ability to predict future events and predict the next activity. They compared the performance of different language models, including GPT-3, BERT, and DeBERTa, and concluded that even relatively small open-source language models have solid outcomes. The authors also found that having detailed, grammatically correct semantic stories matter in the predictive performance. The research also suggests that the predictive performance of SNAP increases with the amount of semantic information within datasets. Finally, the authors indicate that the utilization of robotic process automation technology could enhance data sets and subsequently improve predictions.