Data clean room

A retailer and a brand both want to know whether the brand’s advertising drove the retailer’s sales. Answering it means joining their customer data, and neither can hand the other a customer list.

A clean room is the compromise. Both upload data into an environment neither fully controls. Queries run across the joined set, and only aggregated results come out — never rows. The overlap is computed on hashed identifiers, so the matching happens without either side reading the other’s records.

What it genuinely solves. A measurement question that is otherwise unanswerable, in a form the legal teams on both sides can approve. That last part is most of its value.

The constraints that surprise people.

• Only the overlap is visible. You learn about customers you have in common and nothing about the rest, which is usually the majority.
• Aggregation thresholds hide small groups. Results below a minimum size are suppressed, deliberately, to stop anyone reconstructing an individual by slicing until only one person remains. Narrow segments come back empty.
• You cannot export the join. The answer leaves; the linked data does not. Anything you want to know later has to be asked inside the room.
• The operator matters. Several are run by the same platforms whose advertising is being measured, which is a governance question rather than a technical one.

It is not a way to acquire someone else’s data. It is a way to answer a question over data you will never hold — and the difference is the entire point.