Privacy-Preserving Data Analytics - Collaborating on Data Securely with AWS Clean Rooms
This article explains privacy-preserving data analytics using AWS Clean Rooms. It introduces data collaboration among multiple organizations, access control with analysis rules, and making use of differential privacy.
Privacy Challenges in Data Collaboration
Data sharing between companies is useful for business, but there are many constraints from the perspective of privacy and data protection. Examples include cases where an advertiser and a publisher want to measure ad effectiveness, where a pharmaceutical company and a hospital want to jointly analyze clinical data, and where financial institutions want to share patterns of fraudulent transactions. In such cases, directly sharing raw data is difficult because of privacy regulations (GDPR, the Act on the Protection of Personal Information) and business reasons. AWS Clean Rooms is a service released in 2023 that provides an environment in which multiple organizations can run joint analyses without sharing data. Each participant keeps its own data in S3 and runs queries according to the rules defined within Clean Rooms. Raw data is not exposed to other participants, and only results in the forms permitted by the analysis rules are returned.
Collaborations and Analysis Rules
Using Clean Rooms starts with creating a collaboration. Multiple members (organizations) participate in a collaboration, and each member registers its own data tables as configured tables. You define analysis rules on a configured table and specify the types of queries allowed and the constraints on output. The aggregation rule (Aggregation) allows only aggregate functions such as COUNT, SUM, and AVG and prohibits the output of individual records. You set a lower limit on the number of rows in an aggregation (for example, allowing only aggregations of 100 rows or more) to reduce the risk of identifying individuals from a small number of records. The list rule (List) allows the output of value lists for specific columns, but you can restrict the columns that can be output. The custom rule (Custom) permits running queries from analysis templates that the data owner has verified or queries created by members the owner has permitted, and lets you configure output constraints in detail. With these rules, data owners can control the scope of analysis.
Differential Privacy and Use Cases
The differential privacy option of Clean Rooms adds mathematically calculated noise to query results to make it difficult to infer information about individuals. For example, it adds minute noise to a query result such as "the average purchase amount of men in their 30s living in Tokyo" to make it difficult to back-calculate the purchase amount of a specific individual. You set a privacy budget to limit the number of queries against the same data and reduce information leakage through repeated queries. Main use cases include ad effectiveness measurement (advertisers and publishers analyze conversions without sharing user data) and medical research (multiple medical institutions conduct epidemiological studies without sharing patient data). They also include financial fraud detection (multiple financial institutions detect fraud patterns without sharing transaction data) and retail analytics (manufacturers and retailers forecast demand without sharing sales data).
Technical Mechanisms and Pricing
Clean Rooms references data on S3 directly, and no copying or moving of data occurs. It is integrated with Glue Data Catalog, so you can use existing table definitions as they are. Queries run in an isolated environment within Clean Rooms, and intermediate data is not shared among participants either. When you use the cryptographic computing option, data is processed while it remains encrypted, and even the Clean Rooms service itself cannot access the contents of the data. The pricing for SQL queries in Clean Rooms is determined by CRPU (the unit of Clean Rooms processing) hours. Differential privacy and cryptographic computing each incur additional charges. The cost of storing data is only the S3 charges, and there is no storage charge specific to Clean Rooms.
How to Proceed with Collaboration Design
Data collaboration using Clean Rooms starts with building consensus among the participants. The parties agree in advance on which data to bring together, what kinds of analysis to permit, and what to obtain as results. By using analysis rules to restrict the types of queries that can be run and the content that can be output as results, you can perform only the agreed scope of analysis without disclosing each other's raw data. You also take care with the granularity of the output so that one participant cannot unintentionally infer the other's data in detail. You carefully design the scope of the data provided and the constraints on analysis, and put in place a framework that all participants can agree on.
Summary - Guidelines for Making Use of Clean Rooms
AWS Clean Rooms is a service that achieves privacy-preserving data analytics among multiple organizations. Its main features are query control with analysis rules, mathematical privacy protection with differential privacy, and data protection with cryptographic computing. Because you can gain insights without sharing raw data, you can use it for data collaboration in industries with strict privacy regulations (healthcare, finance, advertising). When you need to share data with partner companies but privacy is a concern, consider Clean Rooms.
References (Official AWS Resources)
The primary sources for this page are the official AWS website and documentation. Check the official pages below for the latest specifications and pricing.
- AWS Clean Rooms official page
- AWS Glue official page
- Amazon S3 official page
- AWS Documentation (official)
If this page and the official documentation disagree, treat the official documentation as authoritative.