Trade-Offs in Leveraging External Data Capabilities: Evidence from a Field Experiment in an Online Search Market
Removing access to an external search engine's autocomplete API reduced suggestion click-through by 4.6%, quantifying the value of external data capabilities.
Summary
The authors ran a 108-day randomised experiment on the mobile app of a large Chinese technology company (about 100 million monthly users), which supplements its own search suggestions with candidates from the market-leading search engine's autocomplete API. Cutting off the API cut click-through on suggestions by 4.6% on average, with heavy users and city dwellers hit hardest, and the effect shrank from about 8 to 9% in the first weeks to about 4% by the end, so short experiments overstate the value of the external capability roughly twofold. The shrinkage came from the firm's own data building up, not from users adapting: reliance on the leader's data slows the accumulation of the firm's internal data and so limits how far its own predictions improve. Downgrading the external candidates instead of removing them halved most of the loss, pointing to a middle path that keeps short-term quality while allowing internal learning. Data-sharing mandates for dominant platforms are therefore not a cure-all for concentrated markets.
Editor's opinion
Borrowing a leader's data capability improves your product immediately but can stunt your own: the more a firm leans on an outside data source, the less it builds the data that would make it independent. The choice is a trade-off between a quality gain now and capability later, and a hybrid that limits the outside source's weight may keep both.
Key concepts
external data capabilities · field experiment · search · API