Data Clean Room for a Retail Media Network or Ad Platform: Use the One Your Cloud Offers, License a Product, or Build Your Own

A data clean room is a shared workspace, usually in the cloud, where two companies match their customer data and run only the analyses they agreed on. Set up well, neither side sees the other's raw records. A retail media network can use one to tell a brand how many shoppers saw its ads and then bought its products. The brand gets a total, not a list of names. You can use the clean room your cloud or data warehouse offers, license a clean-room product, or build your own. The choice depends on where your data and your advertisers' data sit, and whether the clean room is part of what you sell.

What does a data clean room do for a retail media network or an ad platform?

In a typical retail media case, a brand has run sponsored product ads, the paid listings in your search results. You bring ad views and sales, and the brand brings its customer list. Both sides turn each email into a code with the same one-way formula, called hashing, so the same email always gives the same code. The clean room matches the codes, and the brand gets totals, such as how many matched shoppers saw the ads and how many bought.

Staff of the US Federal Trade Commission (FTC) wrote about clean rooms in a November 2024 post on the agency's technology blog. They described them as “cloud data processing services that let companies exchange and analyze data, restrained by rules that limit data use.”

The Interactive Advertising Bureau (IAB), the US trade body for digital advertising, published a framework on clean rooms in commerce on 29 September 2026. Its page lists uses such as “audience overlap, data enrichment, targeting, suppression”. In plain terms: how many customers you share, what else you learn about them, and whom to show ads to or leave out.

Publishers and ad platforms use the same setup. A publisher can match its logged-in readers with an advertiser's customers. A DSP (demand-side platform), the system advertisers use to buy ads, can match the people who saw its ads with an advertiser's sales.

What does a clean room not solve?

A clean room protects data only as well as it is set up. The FTC staff wrote in the same post: “By default, most services that provide DCRs are not privacy preserving.” DCR is short for data clean room. They went on: “Instead, companies must intentionally configure and deploy each constraint for it to be effective.”

Rules on queries do not catch every trick. Google's documentation for BigQuery data clean rooms warns that analysis rules “might not fully prevent sophisticated queries from extracting unauthorized data.” AWS Clean Rooms documentation describes one such trick: learning about one person “by adding or removing records about an individual and observing the difference in query results.”

Hashing does not make data anonymous. FTC staff wrote on the same blog in July 2024: “Companies should not act or claim as if hashing personal information renders it anonymized.” The European Commission says personal data that has been “de-identified, encrypted or pseudonymised but can be used to re-identify a person remains personal data”. So the GDPR, the EU's main data protection law, applies to it. Pseudonymised means names were replaced with codes that can still be linked back to people.

The legal duties stay with each company. IAB Tech Lab writes technical standards for digital advertising. Its clean room guidance calls compliance with privacy law “the responsibility of the Data Contributor”, meaning the company that brings the data. It says partners must trust each other to bring “complete and accurate (not spoofed/doctored/inflated) data”. It also notes that match rates, the share of records that find a pair on the other side, “may be different than expected”.

Sometimes you do not need a clean room at all. The IAB's commerce framework says clean rooms “aren't the right solution for every data or measurement need.”

All sources in this article were opened on 3 November 2026. The FTC quotes are from staff blog posts, not FTC rules. On privacy law, the article quotes only EU and US sources. Other countries have their own rules.

Which clouds and data warehouses offer a clean room?

AWS, Google Cloud, Snowflake and Databricks each offer one inside their own service. A data warehouse is the database where a company keeps its records for analysis. How they differ for a buyer:

• AWS Clean Rooms: AWS says collaboration “doesn't require moving data out of AWS or loading it into another cloud services provider”. The service can also read data kept in Snowflake and Databricks. Whoever sets up the clean room chooses which partner pays for the queries.
• BigQuery data clean rooms, in Google Cloud: Google says its egress controls, the rules on what may leave the room, automatically stop partners who query the data from copying and exporting raw data.
• Snowflake Data Clean Rooms: Snowflake's documentation says “Data Providers must use Snowflake Enterprise Edition” and “The analysis runner bears the cost of the analysis”, so the partner who runs a query pays for it.
• Databricks Clean Rooms: Databricks says collaborators “cannot see the data in other collaborators' tables, views, or volumes”. By default, the other collaborators must approve each piece of analysis code before it runs.

When should we use the clean room our cloud or data warehouse offers?

Use it when most of these are true:

• Your ad logs and sales records already sit on that platform
• Most brands you want to work with are on the same platform, or it can read their data where it sits
• Brands want a few standard analyses, such as audience overlap and sales after a campaign
• You have data engineers to set the rules, write the approved queries and help each new brand start

The catch is brands on other platforms, who may have to open an account on yours or copy their data there. IAB Tech Lab's guidance warns that “Moving data between some DCR platforms may be cumbersome or costly.” The cloud's clean room is also a general tool, not a retail media product. Screens and reports for your brands, and help for each new brand, are yours to build or handle by hand.

When should we license a clean-room product?

License one when:

• The brands you want to work with keep their data on many different clouds and warehouses, and you need most of them this year
• You want ready-made analyses and screens that a brand's marketing team can use without writing code
• Your data team is small, and you would rather pay a vendor to keep the connections working

Some products connect to many platforms at once. LiveRamp's getting-started guide for retailers says you can set up data connections to most cloud-based storage locations, “including AWS, GCS, Azure Blob, Snowflake, Google BigQuery, and Databricks”. It lists a ready-made offering, Retail Intelligence, that “Connects a brand and a RMN (plus optional publisher overlay)”. RMN means retail media network.

The cost is fees and less control over what you can measure. IAB Tech Lab's guidance says “DCR pricing structure can vary dramatically” and suggests asking whether every participant needs its own licence. If you leave later, you may have to rebuild the analyses and set up every brand again.

When is building our own clean room worth it?

Building pays off in these cases:

• The clean room is part of what you sell, and you want your own rules for measurement and audiences
• Audiences must reach your sponsored product auction and your ad server, the system that picks and counts ads, without lists of shoppers leaving your systems
• Fees per query or per partner have grown larger than the cost of a team that runs your own system

Building does not have to mean a new privacy engine. Amazon Ads calls Amazon Marketing Cloud, its clean room for advertisers, a “cloud-based clean room solution” and says it is “Built on AWS Clean Rooms”. Your own layer on top of a cloud clean room can hold the matching, the approved queries, the screens for brands, the logs and the link to your ad server.

The work continues after launch: someone has to review new queries, answer brands' questions about the numbers, apply deletion requests and follow new standards. If nobody on your team will own that, use or license a clean room first.

What do we decide ourselves, whichever path we choose?

The software enforces these rules, but your team decides them:

• Identity matching: what you match people on, such as email or loyalty card number, and how each side prepares it. IAB Tech Lab's PAIR protocol (Publisher Advertiser Identity Reconciliation) “relies on commutative cryptography”. Each side scrambles the identifiers with its own secret key, and the twice-scrambled values can be compared without anyone unscrambling them. PAIR replaced the earlier OPJA specification, which IAB Tech Lab deprecated in July 2025. For measuring which sales followed which ads, IAB Tech Lab finalized ADMaP (Attribution Data Matching Protocol) in February 2025.
• Consent: a record of what each shopper agreed to. IAB Tech Lab's guidance asks every company that brings data to “provide evidence of consent, where applicable/required”.
• Allowed queries: the analyses each brand may run, written and approved in advance. AWS calls them analysis templates, and Google calls them query templates.
• Minimum audience size: each row of a result must cover enough people that no one stands out. IAB Tech Lab's guidance sets no number. Google's Ads Data Hub, its tool for analyzing Google ad data, asks for about 50 unique users per result row in its difference checks mode. It asks for about 20 in its noise injection mode, and about 10 when a query reads only click and conversion data. AWS drops rows below the minimum, and “The query output doesn't indicate that results are removed because of the output constraint.” Tell brands why totals may not add up.
• Noise and query limits: differential privacy adds a little random noise to each answer, so the result barely changes whether one person is in the data or not. On AWS, each query uses up part of a privacy budget. The budget and the noise added per query limit how many queries a partner can run.
• Logs: who ran which query, and when. IAB Tech Lab's guidance says these records “should be transparently visible to all Data Contributors”.
• Getting results out: an audience must reach your ad server or a brand's DSP in a form the ad system can use, not as a downloadable list of emails.
• Deletion: when a collaboration ends, the guidance asks for “prompt removal and destruction of Data Contributor data, as well as any DCR results”.

What should we check before signing with a cloud or a vendor?

• Does our data stay in our own account, or is a copy made somewhere else?
• Does each brand need an account or a licence on the same platform, and who pays for it?
• Who pays for each query, and how does the bill change when the number of brands doubles?
• Can we add our own approved queries and set the minimum audience size, the noise and the query limits for each brand?
• Can we see every query run on our data, with who ran it and when?
• How are deletion requests and withdrawn consent applied to data already in the clean room?
• What can we take with us if we leave: the approved queries, the logs and the match tables that link our shopper IDs to partners' IDs?

Who owns the code, the queries and the data?

On a cloud clean room, the provider owns the service, and the queries your team writes are yours. On a licensed product, analyses written for you are yours only if the contract says so. With your own build, you can own the source code and hire any team to change it. Ask the builder to list in the contract any components it keeps, and the licence that lets you keep using them.

Whatever the path, put three things in the contract. Your shopper and sales data stay yours, and the provider may use them only to run your clean room. The approved queries hold your measurement method, so you should be able to take them with you. Treat the match tables as personal data, and have the contract say where they are kept and when they are deleted.

What should we ask a firm that would build it?

• Will you build on our cloud's clean room or write the core yourselves, and why?
• What will you match shoppers on, and will you follow PAIR for matching and ADMaP for measurement?
• How will each brand's queries be approved, and where is the minimum audience size enforced?
• How will you stop a brand from learning about one shopper by comparing two results?
• How do audiences reach our ad server and our auction, and how long does that take?
• What do the logs record, who can read them, and how long are they kept?
• How will a deletion request or withdrawn consent reach data that has already been matched?
• Can you show this working in a system that runs today?

Where does amBrain fit?

amBrain publishes this blog. Two of its lines describe the systems on either side of a clean room: the ad events going in, and the ad server that uses the audiences coming out. One reads: “amBrain builds event analytics pipelines for ad tech: collection, processing and reporting of impression and click events.” The other: “amBrain builds ad servers: targeting, frequency capping and reporting.” A third line reads: “RTBBidder, a demand-side platform we built for a client.”

amBrain has been building software since 2019. On how it works with clients, it says: “Three formats: full delivery, a dedicated team, or engineers embedded in your team.” Its line on ownership is “The client keeps full ownership of the product and the code, except our reusable components.” Ask for the list of those components by name before you sign.

This article is not a case study. Apart from naming RTBBidder, it describes no client work. It does not say that amBrain has built, licensed or set up a data clean room for any client, and it gives no amBrain prices or timelines. It quotes the other organizations named here from their published pages and claims no relationship between amBrain and any of them.

It is not legal advice either. Check how privacy law applies to your data with a lawyer. If amBrain is on your shortlist, ask it the same questions as any other firm.