Karrot Bets on Local AI to Compete in Secondhand and Community Commerce
Karrot, a local commerce platform for secondhand trading and community services, is making AI central to its product strategy. The company announced this week that its machine learning research on local recommendation systems has been accepted to SIGIR 2026, a top-tier information retrieval conference, while simultaneously launching internal programs to embed AI across product teams.
For product developers, the moves signal a shift in how local platforms can compete against national marketplaces. Rather than broad recommendations, Karrot's system generates suggestions from users within roughly 5 km-a constraint that requires different technical approaches than standard e-commerce.
From Click Rates to Behavioral Embeddings
Karrot spent five years refining its recommendation engine. Early versions tracked click-through rates. The current system uses contrastive-learning embeddings and value functions that measure dwell time and chat initiation-signals that indicate genuine buyer or seller intent.
This progression matters for product teams because it shows how metrics shape recommendations. Measuring engagement differently changes what gets surfaced to users.
Building AI Knowledge Across Teams
Karrot formalized an "AI Show & Tell" relay where employees across functions share AI use cases, experimental results, and lessons learned. The company positions this as a way to speed feature iteration and improve execution quality by distributing AI knowledge beyond specialist teams.
For product organizations, this approach addresses a practical problem: how do you move from isolated AI projects to AI-informed decisions across product, design, and operations?
Scaling Toward Foundation Models
Karrot plans to train foundation-model-level systems on its local-market data. If successful, this could improve user retention, transaction volume, and ad efficiency-metrics that directly affect platform economics.
The company is actively hiring machine learning and AI talent, indicating sustained investment in in-house capabilities. This will increase near-term R&D costs but supports a strategy of building defensible advantages through local-specific models rather than relying on off-the-shelf systems.
For product teams evaluating AI investments, Karrot's approach illustrates a trade-off: specialized, in-house models require sustained hiring and research budgets, but they can generate competitive advantages in specific markets where general-purpose systems fall short.
Learn more about AI for Product Development and Generative AI and LLM approaches for your team.
Your membership also unlocks:
