PostHog uses decision models to cut LLM calls in session-recording pipeline
PostHog has deployed small decision models, such as Jev, alongside large language models to handle simpler classification tasks. In its Replay Vision product, the models run two yes/no checks per session summary to detect personal information, letting 98.7% of summaries skip a costlier Gemini LLM call entirely. PostHog says this approach is up to 444.6x cheaper and 193.6x faster than using frontier LLMs directly.
GoKawiil's interpretation of the reporting above, not reported fact.
The approach suggests a broader industry pattern of pairing cheap, calibrated decision models with expensive LLMs to control inference costs, using LLMs only where nuanced generation is actually needed. Making uncertainty explicit through probability scores could help teams catch and correct errors more systematically than relying on confident but sometimes wrong LLM prose. However, PostHog itself notes these models still make mistakes, implying this is a cost-saving optimization rather than a guaranteed accuracy improvement.
- PostHog combines decision models with LLMs to reduce unnecessary expensive calls
- Two narrow yes/no questions per summary cut false flags to about a tenth of using one broad question
- 98.7% of session summaries bypass the Gemini LLM call entirely in production
Source: newsletter.posthog.com — Jina Yoon, 2026-11-17
Published there as: “Not Everything Needs an LLM”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.