A clean room is a governed computing environment where two or more companies bring separate datasets together, run agreed analysis across the combination, and take out only results. Raw records never move to the counterparty. In advertising the term almost always means a data clean room, abbreviated DCR. The problem it solves is commercial: a brand holds customer records, a media owner holds exposure logs, and the overlap between those files is what shows whether a campaign worked. Neither side will hand the file over. In the European Union and a growing list of US states, neither legally can.

What happens inside one

The workflow runs in four stages, with vocabulary that holds across vendors.

Ingestion comes first. Each participant loads data into a store the operator governs. Identifiers are rarely uploaded in the clear: SHA-256 hashing of email addresses and phone numbers is the baseline, applied after stripping whitespace, lowercasing, and standardizing formats. Inconsistent normalization is the most common cause of a collapsed match rate.

Matching establishes the join key. A direct hash match works when both parties hold the same identifier type. A pseudonymous identity spine, such as LiveRamp's RampID or Unified ID 2.0, resolves different identifiers to a common token first. Or a cryptographic protocol matches without either side holding a comparable value. PAIR, the Publisher Advertiser Identity Reconciliation protocol, uses commutative encryption: each party encrypts its identifiers with its own key, then encrypts the counterparty's already-encrypted output, and because the operations commute, identical inputs produce identical double-encrypted values without decryption. That is why IAB Tech Lab treats PAIR as an interoperability standard rather than a matching convenience: it lets clean rooms run by different companies match against each other.

Analysis is usually SQL, run over event-level tables of impressions, clicks, conversions, and retail or content signals. The analyst writes the joins; the environment decides what comes back.

Output is either an aggregated table landing in a destination the participant controls, such as a BigQuery project, or a modeled segment pushed to an activation endpoint. When New York Times Advertising worked with Decentriq in late 2025, the lookalike segments were exported into Google Ad Manager, so the advertiser needed no presence inside the publisher's ad stack.

The constraints are the product

What separates a clean room from an ordinary data transfer is the set of rules limiting what analysis can run and what can leave. The Federal Trade Commission put it plainly in a November 2024 staff post: the constraints are the differentiator, and only when they are designed, implemented, and monitored properly do they limit use and disclosure.

Those constraints are public. According to Google's Ads Data Hub documentation, the platform applies static checks that reject query patterns before execution, aggregation checks requiring roughly 50 unique users per result row, and difference checks that compare a job against previous results to block re-identification through repeated querying across near-identical date ranges. Queries touching only clicks and conversions drop to about 10 users. Noise injection, which randomizes aggregating SELECT clauses, replaces difference checks and lowers the floor to roughly 20. Failing rows are dropped silently; a filtered row summary is the only way to see how much was excluded. Data access budgets cap how often a dataset can be queried.

Amazon Marketing Cloud reaches a similar result differently. According to Amazon Ads, it accepts only pseudonymized information, advertiser signals cannot be exported or accessed by Amazon, and outputs are aggregated. The user_id column can be used in JOIN, WHERE, and GROUP BY clauses but never selected into a result set. Practitioner documentation describes tiered thresholds varying by column, with commonly cited minimums near 100 unique users; Amazon publishes no single universal figure. ADMaP, finalized by IAB Tech Lab in February 2025, pairs private set intersection with Trusted Execution Environments.

Origin and evolution

Clean rooms began inside walled gardens, replacing data the platforms stopped shipping out.

Google announced the Ads Data Hub beta on 24 May 2017, extending a YouTube measurement project to the Google Display Network and DoubleClick. The context was the removal of log-level DoubleClick IDs from data transfer files, which stripped advertisers of the cross-device attribution they had built on. YouTube then closed the pixel route: Nielsen, Comscore, DoubleVerify, Dynata, Kantar, and Integral Ad Science lost direct measurement access, and from 1 June 2020 campaigns without Ads Data Hub linking faced disruption. Ads Data Hub was not a product advertisers asked for. It was the only remaining path.

Amazon Marketing Cloud launched in beta on 5 January 2021, and Amazon generalized the capability soon after: AWS Clean Rooms, announced at re:Invent on 29 November 2022, reached general availability on 21 March 2023 across eleven regions. Amazon Publisher Cloud, the sell-side application built on it, became generally available in June 2024. The independent tier consolidated fast: Snowflake acquired Samooha in December 2023, BigQuery data clean rooms shipped in April 2024, and LiveRamp bought Habu for $200 million in January 2024.

Standards arrived alongside, and their dating is contested. IAB Tech Lab published its Data Clean Rooms Guidance with the Open Private Join and Activation specification. AdExchanger reported the release on 5 July 2023 and the PDF carries that date, while IAB Tech Lab's own page dates version 1.0 to July 2024. OPJA was deprecated in July 2025, superseded by PAIR, which Google built in its Display and Video 360 team in October 2022 and later donated. PAIR reached version 1.0 in January 2025 and version 1.1 in July 2025.

Why it matters for the marketing community

Clean rooms are now the default interface to the largest platforms' measurement data, which makes competence with them a condition of buying at scale. IAB Tech Lab has positioned PAIR as the unified standard for matching advertiser and publisher first-party data. The IAB State of Data 2023 report found 64% of companies using privacy-enhancing technologies were using clean rooms, and IAB Europe research released on 26 November 2025 found more than half of surveyed European organizations adopting or testing them, with 68% naming cross-platform data access the top challenge.

Outcome data is thin but accumulating. The Decentriq case study recorded a 25% overlap between the publisher's subscriber base and the advertiser's customer file, a 20% uplift in Quality Visit Rate, and a 61% international click-through rate lift against control. Warner Bros. Discovery reported that clean room onboarding produced 3.2 times more unique targetable IDs. Both sets of figures are supplier-published.

Limitations and disputes

The sharpest criticism came from the regulator. In a staff post of 13 November 2024, the FTC wrote that clean rooms "are not rooms, do not clean data, and have complicated implications for user privacy." Protections are not automatic; each constraint must be deliberately configured, and services often default to allowing both parties full access, which makes misconfiguration expensive. By letting more companies exchange more data, clean rooms also expand the perimeter that must be defended, the agency citing a service without two-factor authentication as an example of how one credential attack could compromise millions of records.

Operationally, the aggregation floors that make clean rooms defensible also make them blind. Small segments, narrow geographies, and niche categories fall below thresholds and vanish without an error message. Skill is the second barrier, which is why Amazon's move to open Marketing Cloud to sponsored ads advertisers on 16 September 2025 with no-code templates mattered.

Neutrality is the newest dispute. Publicis Groupe agreed on 17 May 2026 to acquire LiveRamp for $2.5 billion at $38.50 per share, folding a large independent operator into a holding company that is also a major customer of the publishers routing data through it. A June 2026 SEC filing described the safeguards in architectural terms, citing Trusted Execution Environments and clean room firewalls rather than policy.

Disambiguation

A cleanroom in manufacturing is a physical space with controlled particulate counts, used in semiconductor and pharmaceutical production. Clean room design in software means reimplementing a specification without exposure to the original code. Neither shares a mechanism with the advertising sense.

A walled garden is a market position, not a technology. Ads Data Hub and Amazon Marketing Cloud are clean rooms operated by walled gardens; AWS Clean Rooms, Snowflake, Decentriq, and InfoSum are sold as neutral infrastructure. The category describes governance, not ownership.

An identity graph resolves scattered identifiers to a common person or household token; a clean room is the governed environment where the join happens. Most collaborations use both, and confusing them obscures where a match rate is won or lost.

A customer data platform unifies one company's own data for its own use. A clean room handles data belonging to someone else.

Recent developments

Standards work has moved from matching to plumbing. IAB Tech Lab published ECAPI Guidelines for Data Clean Rooms with comments closing on 4 September 2026, covering how conversion events should be packaged for upload: Parquet Modular Encryption or application-layer encryption for values the core standard designates as hashed, Trusted Execution Environments with key-encryption-key and data-encryption-key wrapping, and field mappings rendering JSON numbers as DOUBLE. Tech Lab research cited in the document found 72% of publishers naming technical integration a major obstacle to conversion API adoption. The guidelines standardize transmission, not the legal basis for collection.

Compute capacity is the other axis. LiveRamp added NVIDIA GPU infrastructure in April 2026, reporting model training at up to 15 times CPU speed, and in August 2026 brought de-duplicated Meta measurement into Cross-Media Intelligence. Amazon folded clean room provisioning into its unified advertiser account on 30 July 2026, and Netflix confirmed at its 2026 upfront that InfoSum would join Snowflake and AWS by year-end.

Timeline

  • 24 May 2017: Google announces the Ads Data Hub beta, extending it beyond YouTube to the Google Display Network and DoubleClick
  • 1 June 2020: YouTube campaigns without Ads Data Hub linking become subject to third-party measurement disruption
  • 5 January 2021: Amazon Marketing Cloud enters beta in North America, Europe, Japan, and Australia
  • October 2022: Google's Display and Video 360 team develops the PAIR protocol
  • 29 November 2022: AWS announces Clean Rooms at re:Invent
  • 12 January 2023: AWS Clean Rooms enters preview
  • February 2023: IAB Tech Lab opens DCR Guidance and the OPJA specification for public comment
  • 21 March 2023: AWS Clean Rooms reaches general availability across eleven regions
  • 5 July 2023: IAB Tech Lab publishes DCR Guidance version 1.0 and OPJA version 1.0, per AdExchanger and the guidance document itself; Tech Lab's website dates version 1.0 to July 2024
  • October 2023: Amazon Publisher Cloud launches on AWS Clean Rooms
  • 29 November 2023: AWS Clean Rooms Differential Privacy enters preview
  • December 2023: Snowflake acquires Samooha
  • January 2024: LiveRamp acquires Habu for $200 million
  • March 2024: Snowflake Data Clean Rooms reach general availability
  • April 2024: BigQuery data clean rooms reach general availability
  • June 2024: Amazon Publisher Cloud becomes generally available for streaming TV and web publishers in the US and Canada
  • 13 November 2024: FTC staff publish "Data Clean Rooms: Separating Fact from Fiction"
  • January 2025: PAIR protocol version 1.0 released
  • February 2025: ADMaP version 1.0 finalized
  • July 2025: PAIR version 1.1 finalized; OPJA deprecated
  • 16 September 2025: Amazon opens Marketing Cloud to sponsored ads advertisers without partner coordination
  • 2 October 2025: NIQ launches a global data clean room on Snowflake
  • 26 November 2025: IAB Europe publishes pan-European addressability research covering 27 markets
  • April 2026: LiveRamp adds NVIDIA GPU infrastructure to its clean rooms
  • 17 May 2026: Publicis Groupe agrees to acquire LiveRamp for $2.5 billion
  • 19 June 2026: Amazon Publisher Cloud introduces Outcome Optimizer for FreeWheel deals
  • 30 July 2026: Amazon provisions clean room access through the unified advertiser account
  • 4 September 2026: Comment period closes on IAB Tech Lab's ECAPI Guidelines for Data Clean Rooms

Summary

Who: Advertisers, publishers, retailers, and measurement vendors on one side; platform operators and infrastructure providers on the other. Walled garden clean rooms include Google Ads Data Hub, Amazon Marketing Cloud, and Meta Advanced Analytics. Cloud and independent operators include AWS Clean Rooms, Snowflake, BigQuery data clean rooms, Databricks, LiveRamp, InfoSum, Decentriq, and Optable. IAB Tech Lab sets the interoperability standards.

What: A governed environment for joint analysis of separate datasets, distinguished from an ordinary data transfer by constraints on what queries may run and what results may leave. Typical controls include hashed or cryptographically matched identifiers, minimum aggregation thresholds per output row, difference checks or noise injection against repeated querying, non-selectable user identifier columns, and Trusted Execution Environments.

When: The category began with Google's Ads Data Hub beta in May 2017 and Amazon Marketing Cloud in January 2021, generalized into cloud services during 2023 and 2024, and acquired formal interoperability standards between 2023 and 2025 through IAB Tech Lab's DCR Guidance, PAIR, and ADMaP.

Where: Inside cloud infrastructure operated by the platform or a vendor. Data typically stays within the participant's own cloud governance boundary, with newer architectures moving the compute to the data rather than the reverse.

Why: Third-party cookie and mobile identifier loss, GDPR and US state privacy law, and platform decisions to stop exporting log-level data left advertisers and publishers unable to measure or match without exposing customer records. Clean rooms exist to produce the overlap without producing the transfer. Whether they protect the people in the datasets, as the FTC noted in November 2024, depends entirely on how the constraints are configured.