Verification
Part of Advertising clean rooms
What a clean room can and cannot reveal
See what clean-room outputs establish, how disclosure controls differ and which populations remain unknown.
A clean-room result is limited by the records that entered, the matches that were possible and the output rules that applied. It may reveal an aggregate overlap or a campaign outcome among matched records. It cannot describe people or events that were never observed or could not be joined. Nor does the name guarantee aggregate-only outputs: some configurations permit overlap lists or activation.
Identify the population behind the result
Suppose a publisher supplies exposure events and an advertiser supplies purchases. Their overlap can exclude records without a compatible key, records outside the agreed dates and records ineligible for the proposed use. A low overlap may reflect any of these limits.
Ask what one row represents: a person, account, device, impression or order. If one customer has several orders or impressions, a simple joined-row count must not be reported as distinct customers.
Where disclosure rules allow it, request separate counts for supplied, eligible, matched and unmatched records. State the denominator for any match rate.
Key population metrics in clean-room analysis
- Supplied records
- Total input data from each party
- Eligible records
- Records meeting agreed criteria (e.g. date range, key compatibility)
- Matched records
- Records successfully joined via shared key
- Unmatched records
- Records excluded due to missing keys or eligibility constraints
Identify what may leave
AWS Clean Rooms has different rule types. Its aggregation rule supports queries that aggregate statistics using COUNT, SUM and AVG across optional dimensions; list and custom rules are also available. Custom rules can permit reviewed queries or queries from authorised accounts.
Snowflake activation can write approved template results to a destination outside the clean room when the template and data-provider permissions allow it.
These are product capabilities, not a description of a particular collaboration. Before running an analysis, confirm the proposed output fields, groupings, recipients and use.
Check whether a join key may be returned, used only for matching or included in an approved activation. A group-size threshold alone does not describe every permitted comparison.
Pre-analysis verification checklist for clean-room collaborations
- Confirm output fields and groupingsEnsure alignment with agreed use case
- Verify join key usage rulesCheck if key is returned, used only for matching, or approved for activation
- Validate recipient and use permissionsEnsure authorised accounts and destinations are specified
- Review output rule limitationsConfirm whether suppression thresholds apply to groups
Say only what the result establishes
An overlap is not the size of the whole customer base. A matched exposure does not prove the person noticed the ad. A purchase after exposure does not prove the ad caused it. Those claims require separate delivery, attention or study-design evidence.
A suppressed segment should be labelled unavailable under the output rule, not zero.
Attach an interpretation note to each result: input population, matching unit, observation period, output rule, suppressed or unknown groups, and the precise question answered.
