Target's hybrid retail search lifts CTR 0.97% and halves zero-result queries
_reachsumit · x · 2026-09-28
Target engineers detail their production hybrid search system in a new paper (arXiv:2609.31498): keyword search runs in parallel with a fine-tuned dual-encoder, attribute filters apply to vector results, and weighted interleaving fuses both lists. Online A/B tests vs lexical-only search showed +0.97% CTR, +0.98% order conversion, and +1.10% demand per visitor, while roughly halving zero-result searches. The paper covers data processing, embedding training, precision control, fusion strategy comparison, and latency optimizations; the system serves millions of customers daily.
More from Apps
- shadcn: Liquid Glass improved in iOS/macOS 27, but UX still regressed — shadcn · 2026-09-28
- One Obsidian Vault for Goals, Projects and Time: Let AI Plan Your Week — dSebastien · 2026-09-28
- User saves $2,130 on car insurance after letting AI agent compare quotes and bind a GEICO policy — armand_ruiz · 2026-09-28
- Cut literature screening time in half with this ResearchRabbit workflow — Faheem_uh · 2026-09-28
- The Blank Box Problem: Consumer AI Still Too Hard for Mainstream Users to Learn — viksit · 2026-09-28
- Anthropic design lead imagines a claude-artifacts-meets-Google-Docs writing tool — jeff_weinstein · 2026-09-28