Case Study: Grubhub turns messy web signals into better restaurant scoring with TinyFish

GrubHub adds 353 best restaurants and expands 20% with TinyFish

TinyFish helped Grubhub overcome a challenge faced by its merchant analytics team. While Grubhub had a strong scoring model, it systematically missed culturally and locally important restaurants because their signals were trapped in fragmented, messy sources like Reddit, YouTube, and local review communities.

The solution involved using TinyFish Agents to continuously traverse discussions and resolve restaurant entities from noisy mentions, turning them into structured outputs. This allowed Grubhub to incorporate these web signals into its merchant scoring system, resulting in the validation of 353 net-new "best" restaurants in Boston and a 20% cohort expansion.

GrubHub