I Fixed Alamo Drafthouse's Search With Query Logs Alone: +18.4% MRR, No Model Training
CShorten30 · x · 2026-11-12
Search engineer softwaredoug rebuilt Alamo Drafthouse's QR-code ordering search as an annoyed customer — typing 'burger' returned nachos. Measured across 2,327 simulated search sessions: +18.4% MRR and 72.4% fewer zero-result sessions.
- Diagnosis: the site implements navigational search (Apple Spotlight-style prefix filtering, 'bur' matches bur), with essentially no relevance ranking and many false positives.
- Core method: treat your site's user search traffic as a corpus — store historical queries and clicks, aggregate what's frequently clicked, and expose that as a search-time signal dataset.
- When it works best: small corpus, decent traffic concentrated in a fat head of repeated queries. Far simpler than fine-tuning rerankers/embeddings, with no model training and no GPUs.
More from Apps
- Ethan Mollick: 'talk to your documents' RAG solutions already feel dated — emollick · 2026-11-12
- Real projects from a ChatGPT Sites hackathon in Singapore — craigsdennis · 2026-11-12
- Adobe embeds generative AI in Premiere and Firefly with a built-in creative agent — emmanuelvivier · 2026-11-12
- IBM and Microsoft launch Imagine Enterprise, governed AI agents for regulated sectors — emmanuelvivier · 2026-11-12
- Google begins rolling out Gemini to Apple CarPlay worldwide — Polymarket · 2026-11-12
- iKnow open-sources a read-only viewer into macOS's hidden personal knowledge graph — JeremyCMorgan · 2026-11-12