This post discusses the use of decision models at PostHog to improve efficiency in tasks that typically require LLM calls. It covers practical use cases, including using decision models to decide whether to make expensive calls, routing models to the most suitable options, handling queries with simple checks, and opting for decision models when LLMs are overkill. The author shares lessons learned, such as the importance of setting the right thresholds to avoid false positives and negatives, and the distinction between simpler and complex tasks suitable for decision models versus LLMs.