In the dynamic landscape of online advertising, precise measurement is key to optimizing ad performance. Advertisers rely on measuring user interactions, like views and clicks, and attributing these to conversions such as purchases or sign-ups. However, increasing privacy concerns and tightening regulations—like GDPR and Apple's App Tracking Transparency—have made user data collection challenging. Balancing accurate measurement with robust privacy protection is now more crucial than ever.

At TikTok Privacy Innovation, we've pioneered a groundbreaking solution: AdsBPC (Ads Measurement with Bounded Per-Day Contributions), a differential privacy-based method specifically tailored for online advertising measurement.

What makes AdsBPC unique?

AdsBPC addresses two major challenges:

• Real-Time Streaming Measurement: Ensuring privacy for continuous, real-time data streams without rapidly exhausting the privacy budget.
• User-Level Privacy: Protecting individual users comprehensively across multiple interactions rather than just isolated events.

How AdsBPC works

AdsBPC introduces a user-level differential privacy algorithm optimized for advertising measurement tasks:

1. Bounded Contributions: Sets daily limits on user contributions to reduce sensitivity, balancing accuracy and privacy.
2. Adaptive Noise Scaling: Dynamically adjusts noise levels based on user behavior data, ensuring optimal accuracy.
3. Streaming Optimization: Supports continuous measurement, allowing advertisers real-time insights into their campaign performance.

Key benefits

• Superior Accuracy: Experiments with synthetic and real-world data have demonstrated that AdsBPC improves accuracy by approximately 33% to 95% compared to previous methods.
• Flexible and Adaptive: Easily adapts to various campaign goals, requiring minimal structural changes.
• User Trust and Compliance: Enhanced formal privacy through rigorous mathematical definitions, aligning with global privacy regulations.

Practical applications and impact

Using AdsBPC, advertisers can confidently perform the following:

• Track conversion trends over campaigns in real-time.
• Conduct detailed analyses (for example, prefix-sum and sliding window queries) with minimal loss in data accuracy.
• Make informed, privacy-preserving decisions that enhance ad performance and consumer trust.

Why it matters

Privacy concerns have transformed the digital advertising landscape. AdsBPC not only helps advertisers navigate this landscape effectively but also validates that user privacy isn't compromised in pursuit of advertising insights.

Join us in advancing privacy-preserving advertising measurement. With AdsBPC, advertisers can achieve precise measurement without compromising user trust.

Workshop presentation

The AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-25) serves as a platform for researchers, AI practitioners, and policymakers to discuss both technical and societal challenges and present innovative solutions related to privacy in AI applications. AdsBPC was presented as a poster at the PPAI-25 workshop, and with its potential for substantial impact, AdsBPC stands to contribute meaningfully to the advancement of privacy-preserving AI.

To learn more, check out the workshop summary and published paper.