Attribution Data Clean Rooms: Privacy-Safe Measurement

Data clean rooms are secure environments where two parties (typically an advertiser and a publisher or ad network) can match their first-party data without either party seeing the other's raw data. In mobile attribution, clean rooms let you measure campaign performance without sharing user-level data with ad networks, and without ad networks sharing their user data with you.

This guide explains how data clean rooms work for mobile attribution. For attribution and privacy regulations, see mobile attribution under GDPR. For SKAdNetwork, see SKAdNetwork and SKAN 5.0: complete developer guide.

Why Data Clean Rooms Exist

Traditional mobile attribution works by sharing user-level data between parties:

1. Ad network shares click data (device ID, timestamp, campaign).
2. Attribution provider matches clicks to installs using device IDs.
3. Results are shared back to the ad network.

This model requires both parties to share raw user data, which conflicts with privacy regulations (GDPR, CCPA) and platform restrictions (ATT, Privacy Sandbox).

Data clean rooms solve this by providing a neutral environment where matching happens without data exposure.

How Clean Rooms Work

The Basic Flow

1. Advertiser uploads first-party data (user IDs, conversion events)
   → Data is encrypted/hashed before upload

2. Publisher/Ad Network uploads campaign data (impressions, clicks, user IDs)
   → Data is encrypted/hashed before upload

3. Clean room matches records using hashed identifiers
   → Neither party sees the other's raw data

4. Clean room outputs aggregate results
   → "Campaign X drove 1,200 installs with $45,000 revenue"
   → No user-level data is exposed

Privacy Protections

Clean rooms enforce privacy through several mechanisms:

Protection	How It Works
Data isolation	Each party's data is stored separately and encrypted
No raw data export	Neither party can export the other's raw data
Aggregate-only output	Results are aggregated (no individual user records)
Minimum thresholds	Results with fewer than a minimum number of users (e.g., 50) are suppressed
Query restrictions	Only pre-approved queries are allowed (no arbitrary SQL)
Audit logging	All queries and access are logged for compliance

Clean Room Providers

Major Platforms

Provider	Type	Strengths
Google Ads Data Hub	Walled garden	Access to Google's ad data (Search, YouTube, Display)
Meta Advanced Analytics	Walled garden	Access to Meta's ad data (Facebook, Instagram)
AWS Clean Rooms	Infrastructure	Flexible, use with any data partner
Snowflake Data Clean Rooms	Infrastructure	SQL-based, integrates with existing Snowflake data
InfoSum	Independent	Decentralized (data never leaves your infrastructure)
LiveRamp Data Collaboration	Independent	Identity resolution + clean room

Walled Garden vs. Infrastructure Clean Rooms

Walled garden clean rooms (Google, Meta) are built by the ad platform itself. You bring your conversion data; they have the ad exposure data. You can only measure campaigns on that specific platform.

Infrastructure clean rooms (AWS, Snowflake) provide the technology, and you choose your data partners. More flexible but require more setup.

Clean Rooms for Mobile Attribution

Measuring Install Campaigns

Instead of sharing device IDs with an ad network for real-time attribution, you can use a clean room for aggregate measurement:

-- Example clean room query (aggregate results only)
SELECT
  campaign_id,
  ad_group_id,
  COUNT(DISTINCT matched_users) AS attributed_installs,
  SUM(revenue_30d) AS revenue,
  SUM(revenue_30d) / COUNT(DISTINCT matched_users) AS revenue_per_install
FROM
  advertiser_installs
JOIN
  publisher_impressions
ON
  hashed_user_id = hashed_user_id
WHERE
  impression_time < install_time
  AND install_time - impression_time < INTERVAL '7 days'
GROUP BY
  campaign_id, ad_group_id
HAVING
  COUNT(DISTINCT matched_users) >= 50  -- minimum threshold

Comparing to Traditional Attribution

Aspect	Traditional Attribution	Clean Room Attribution
Data sharing	User-level data shared between parties	No raw data shared
Granularity	User-level (real-time)	Aggregate (batch)
Latency	Real-time (seconds to minutes)	Batch (hours to days)
Campaign optimization	Real-time bidding optimization	Post-campaign analysis
Privacy compliance	Requires consent for data sharing	Privacy-safe by design
Cost	Included in attribution provider fee	Separate cost (query fees, infrastructure)
Complexity	Low (SDK integration)	High (data engineering required)

When Clean Rooms Make Sense

Good Use Cases

• Measuring walled garden campaigns. Google and Meta campaigns where you want deeper analysis than the standard dashboard provides.
• Cross-publisher measurement. Comparing performance across multiple ad networks without sharing data between them.
• Privacy-regulated markets. EU markets where GDPR restricts user-level data sharing.
• Post-ATT iOS measurement. Supplementing SKAdNetwork data with aggregate clean room analysis.
• High-spend campaigns. When campaign budgets justify the cost and complexity of clean room setup.

When Clean Rooms Are Overkill

• Small ad budgets. If you spend less than $50,000/month on mobile ads, the standard attribution provider reports are sufficient.
• Real-time optimization. Clean rooms provide batch results, not real-time signals for bidding.
• Simple attribution needs. If last-click attribution from your provider meets your needs, clean rooms add complexity without proportional value.

Implementation Considerations

Data Preparation

Your first-party data must be prepared before upload:

import hashlib

def prepare_for_clean_room(user_data):
    prepared = []

    for user in user_data:
        record = {
            # Hash PII before upload
            'hashed_email': hashlib.sha256(
                user['email'].lower().strip().encode()
            ).hexdigest(),
            'hashed_phone': hashlib.sha256(
                user['phone'].strip().encode()
            ).hexdigest() if user.get('phone') else None,

            # Non-PII conversion data (sent as-is)
            'install_date': user['install_date'],
            'revenue_7d': user['revenue_7d'],
            'revenue_30d': user['revenue_30d'],
            'subscription_status': user['subscription_status'],
            'country': user['country']
        }
        prepared.append(record)

    return prepared

Match Rates

Clean rooms match users based on shared identifiers (typically hashed email or phone). Match rates vary:

Identifier	Typical Match Rate
Hashed email	30-60% (depends on whether users signed up with the same email on both platforms)
Hashed phone	20-40%
MAID (device ID)	Declining (ATT, Privacy Sandbox)
Combined (email + phone)	40-70%

A 50% match rate means half of your users cannot be measured through the clean room. Factor this into your analysis.

Cost

Clean room costs include:

• Infrastructure fees. AWS Clean Rooms charges per query. Snowflake charges per compute credit.
• Data preparation. Engineering time to format and upload data.
• Analysis. Data science time to write queries and interpret results.
• Ongoing maintenance. Regular data uploads and query updates.

Expect $5,000-$50,000/month depending on scale and complexity.

Tolinku for Privacy-Safe Attribution

Tolinku's analytics provide deep link attribution without requiring user-level data sharing with third parties. Click and conversion data stays in your Tolinku dashboard. Configure privacy-safe tracking in the Tolinku dashboard.

For mobile attribution, see mobile attribution: a developer's guide. For GDPR compliance, see mobile attribution under GDPR.