How Alibaba’s Secure Data Hub Powers Cross‑Domain Advertising Tracking with Privacy‑Preserving Computation

Overview

Alibaba Mama’s Secure Data Hub (SDH) is a Data Clean Room platform that enables advertisers, DSP/DMP partners, and other ecosystem players to fuse data, perform privacy‑preserving computation, and build joint models while keeping raw data invisible.

Background

Industry Context

Advertising generates massive, fragmented data across many roles (users, media, advertisers, SSP, ADX, DSP, DMP, CDP). This fragmentation creates data islands that hinder efficient data sharing and compliance.

Privacy‑enhancing computation (PEC) is essential to unlock data value while protecting user privacy. Four core PEC techniques are used in advertising:

Secure Multi‑Party Computation (MPC) : secret sharing, garbled circuits, homomorphic encryption enable joint computation without exposing raw inputs.

Federated Learning (FL) : distributed model training that keeps raw samples on‑device, exchanging only model parameters protected by privacy‑preserving computation (PPC).

Trusted Execution Environment (TEE) : hardware‑based enclaves that isolate data and code during joint analysis.

Data Clean Room (DCR) : a unified environment that combines the above techniques to provide data isolation, authentication, and differential‑privacy guarantees.

Collaboration Background

Alibaba Mama partnered with Jihe Technology (ReachMax) to apply SDH for cross‑domain advertising tracking and full‑domain asset analysis, leveraging each party’s ad and user data.

Use‑Case 1: Cross‑Domain Advertising Tracking

Business Scenario

Advertisers run campaigns on multiple platforms; user interactions are scattered across channels and landing pages (e.g., Taobao/Tmall stores, brand websites, mini‑programs).

Business Requirements

Track user journeys across channels, attribute conversions, and provide end‑to‑end measurement.

Solution

Participants : Alibaba Mama (platform side) and brand advertisers using ReachMax.

Data Distribution : Platform ad‑exposure data, ReachMax public‑domain ad data, and brand‑private user data.

Privacy Requirements : Data never leaves its domain; only aggregated results are exposed with added noise.

Privacy Techniques : MPC, PSI, PSU, Differential Privacy (DP).

Process Flow :

PSI matches ReachMax device‑ID ad data with Alibaba Mama’s user‑ID mapping to identify cross‑domain users.

PSU combines PSI results with platform ad‑exposure data for attribution calculations.

MPC performs joint attribution analysis with e‑commerce conversion data.

DP adds noise to the final attribution results.

Aggregated metrics are released per dimension.

Results : Generated reports covering five dimensions (channel, media, placement, creative, audience) and 20 key performance indicators (e.g., views, clicks, adds‑to‑cart, purchases). MPC tasks on billion‑scale data completed in 20‑30 minutes.

Use‑Case 2: Full‑Domain Asset Analysis

Business Scenario

Advertisers need a holistic view of user assets spread across multiple ad platforms and media channels.

Solution

Participants : Alibaba Mama and ReachMax brand advertisers.

Data Distribution : Public‑domain ad data, brand‑private data, and Alibaba Mama’s e‑commerce conversion and tag data.

Privacy Requirements : Same as above – no data leaves its domain, only aggregated, noise‑protected results are shared.

Privacy Techniques : MPC, PSI, DP.

Process Flow :

PSI links ReachMax device‑ID ad data with brand‑private data for cross‑domain user identification.

MPC analyzes the combined dataset to produce audience portraits (gender, age, purchase potential, city tier, strategic segments).

DP adds noise to protect individual records.

Aggregated portrait and performance reports are released.

Results : Delivered audience segmentation reports and standard performance dashboards covering the same 20 KPIs; PSI+MPC tasks on billion‑scale data completed in 40‑50 minutes.

Benefits & Impact

Data Security : MPC ensures data never leaves its private domain; fine‑grained metadata controls and end‑to‑end encryption protect confidentiality.

Analysis Diversity : SDH offers a rich set of components for audience insight, attribution, and efficiency measurement across multiple dimensions.

Accuracy : MPC achieves a calculation accuracy of 2⁻³², delivering precise statistical results.

Timeliness : Reports are generated within T+1 day, enabling near‑real‑time optimization and improving analysis efficiency by over 80%.

Future Outlook

SDH will continue to enhance SaaS capabilities, support more complex joint statistics and modeling, improve MPC/FL performance at hundred‑billion scales, and expand privacy‑preserving applications across the advertising workflow (targeting, insight, optimization, attribution, efficiency measurement, reach monitoring).