{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"Chain of Thought | AI Agents, Infrastructure & Engineering","title":"How Intercom Cut $250K/Month by Ditching GPT for Qwen","html":"<iframe width=\"100%\" height=\"180\" frameborder=\"no\" scrolling=\"no\" seamless src=\"https://share.transistor.fm/e/0fd18337\"></iframe>","width":"100%","height":180,"duration":3211,"description":"Intercom was spending $250K/month on a single summarization task using GPT. Then they replaced it with a fine-tuned 14B parameter Qwen model and saved almost all of it. In this episode, Intercom's Chief AI Officer, Fergal Reid, walks through exactly how they made that call, where their approach has changed over time, and how all of their efforts built their Fin customer service agent. Fergal breaks down how Fin went from 30% to nearly 70% resolution rate and why most of those gains came from surrounding systems (custom re-rankers, retrieval models, query canonicalization), not the core frontier LLM. He explains why higher latency counterintuitively increases resolution rates, how they built a custom re-ranker that outperformed Cohere using ModernBERT, and why he believes vertically integrated AI products will win in the long term.If you're deciding between fine-tuning open-weight models and using frontier APIs in production, you won't find a more detailed decision process walkthrough.🔗 Connect with Fergal: Twitter/X: https://x.com/fergal_reidLinkedIn: https://www.linkedin.com/in/fergalreid/Fin: https://fin.ai/🔗 Connect with Conor:YouTube: https://www.youtube.com/@ConorBronsdonNewsletter: https://conorbronsdon.substack.com/Twitter/X: https://x.com/ConorBronsdonLinkedIn: https://www.linkedin.com/in/conorbronsdon/🔗 More episodes: https://chainofthought.showCHAPTERS0:00 Intro0:46 Why Intercom Completely Reversed Their Fine-Tuning Position8:00 The $250K/Month Summarization Task (Query Canonicalization)11:25 Training Infrastructure: H200s, LoRA to Full SFT, and GRPO14:09 Why Qwen Models Specifically Work for Production18:03 Goodhart's Law: When Benchmarks Lie19:47 A/B Testing AI in Production: Soft vs. Hard Resolutions25:09 The Latency Paradox: Why Slower Responses Get More Resolutions26:33 Why Per-Customer Prompt Branching Is Technical Debt28:51 Sponsor: Galileo29:36 Hiring Scientists, Not Just Engineers32:15 Context Engineering: Intercom's Full RAG Pipeline35:35...","thumbnail_url":"https://img.transistorcdn.com/opYb_ogIEF0JJPpA1L13NhLPore7cUzGtcL8Okdgfg4/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS81YjBh/ODMzMTY3ZjQ0MjBj/YTE1ODMwYTZlNDgx/Mjc2Mi5qcGc.webp","thumbnail_width":300,"thumbnail_height":300}