How Many People Already Have This Dataset?

A vendor's sales deck never says "500 hedge funds already trade this." But the question of how crowded a dataset already is matters just as much as whether the data is any good, because a genuinely predictive signal loses most of its value once enough capital is chasing the same trades at the same time. Before paying for a new feed, it is worth spending real effort trying to estimate how many other desks already have it.

Indirect ways to estimate crowding

There is rarely a direct count available, so researchers triangulate from proxies: how long the vendor has sold the data and to how many clients (ask directly — some will disclose client counts, if not names); whether the underlying raw data source (e.g., a satellite operator, a card-panel provider) sells to multiple resellers who repackage the same base data differently; how the data behaves around known crowded events, such as unusually fast price reaction on the exact day a similar public data release occurs; and simply how expensive and exclusive the vendor's pricing tier is, since heavily marketed, cheaply priced feeds are sold widely almost by construction.

Worked example

A team is evaluating a card-transaction dataset priced at $180,000 a year. The vendor discloses, when asked directly, that it currently has 40 institutional clients, up from 12 three years ago. Cross-referencing job postings and public conference talks, the team finds at least six large multi-strategy funds have publicly discussed using card-panel data for retail-sector research over the past two years. Combined with the vendor's own growth in client count, the team concludes the dataset's core retail-sales signal is likely well arbitraged for large, liquid names, but may still have room in a smaller-cap subset of tickers that the largest funds are less likely to bother sizing into. They negotiate a lower-tier subscription focused on that subset rather than paying full price for the broad universe.

What this means in practice

Crowding is not a reason to automatically reject a dataset, but it should shift where you look for edge — toward less-covered corners of the universe, faster-decaying, less commoditized derivatives of the raw data, or combinations with proprietary data the vendor's other clients do not have.