On February 19th, 2025, Prebid hosted a webinar titled “ASTEROID: Advanced System for Testing, Experimentation, Research, Optimization of Identity Data,” providing valuable insights into identity-based signal optimization and system efficacy measurement for publishers.

Speakers and Agenda

The session featured key Prebid team members including: 

• Nic Gallardo (VP, Engineering – Freestar)
• Matt Griffiths (SVP, Technology – Audigent)
• Garrett McGrath (SVP, Product Management – Magnite)
• Jeff Wieland (Principal Technical Product Manager – Magnite) 

The agenda covered the background and context of identity signals in targeting, a detailed technical overview, and a case study demonstration.

Background and Context

Garrett initiated the discussion by explaining the degradation of identity-based signals used for targeting, highlighting the changes and challenges faced by the industry in 2024. This led to a review of the ‘Connect’ concept initiated in Q2 2023 and its updates, including the results showing a 13% lift in performance and key learnings from ongoing tests.

The Problem and Solution

Matt Griffiths discussed the inherent complexities in managing and measuring the efficacy of identity systems. The existing manual, hard-to-configure, and non-scalable frameworks posed significant challenges. 

The new approach, introduced during the webinar, aims to democratize and enhance transparency through a standardized, easy-to-configure, scalable solution that extends to analytics providers for improved reach and performance. ASTEROID’s ultimate goal is to give more control to Publishers.

Prebid does not dictate which providers to work with, recognizing that not every mix of ID companies will be effective for every publisher, yet it offers a framework to genuinely test and demonstrate which ones can drive success.

Technical Overview and Benefits

Nic Gallardo provided a technical overview of the new system, which includes an extension of the current modules to manage identity system testing and lift measurement. The framework allows for both randomized and strategic testing, greatly simplifying identity systems testing management and enabling data-driven decisions through insightful analytics. He emphasized the benefits of using randomized testing to reduce selection bias and accurately measure performance across different traffic conditions, providing crucial insights.

Nic also explained ASTEROID’s initial workflow, focusing on how publishers can decide the type of testing—strategic or broad coverage—that suits their needs, such as allocating 20% of traffic to test a new analytics adapter. This setup enables publishers to capture directional data by controlling the distribution of traffic. The integration of ID system configurations within the PBJS bid lifecycle was also explained, illustrating how during the bid event, the configuration is emitted and coupled with the analytics adapter, which then processes and aggregates the data. This systematic approach allows for the data to be effectively interpreted and presented, leveraging the analytics capabilities built in-house or by external providers.

Additionally, Nic talked about Ad Server Reporting to offer a thorough yield analysis. This approach allows publishers to gain detailed insights into how different identity partners influence key performance indicators such as fill rates, viewability, and CTRs. He also highlighted the capability to directly compare the revenue uplift resulting from ASTEROID tests against revenue reported by GAM, enabling a clear evaluation of the impact and efficacy of the implemented strategies.

Regarding Analytics, Nic touched on the integration and implementation of analytics adapters, whether externally contracted or built in-house, to enable real-time data ingestion and later aggregation for comprehensive analysis. This setup allows publishers to work directly with analytics adapter companies to tailor data collection to their specific needs, potentially including new data points and visualizations on their dashboards. The integration leverages the existing bid one events and extends the analytics capabilities without necessitating new APIs or endpoints, thus streamlining data processes and enhancing yield and efficacy measurements through the “ASTEROID” object.

In practice, it looks something like this:

Nic shared a use case for the Analytics to show how different identity systems impact yield under various conditions, such as different geographies or browser environments, thereby guiding strategic decisions on which identity systems to employ for optimal results. However, it was noted that this data should be viewed as directional rather than definitive, acknowledging the variability in individual sessions and the absence of a one-size-fits-all solution in identity systems.

Case Study and Results

Prebid conducted a case study on US traffic with the setup as follows:

In the case study involving a 95/5 split test, identity partners were evaluated based on browser and device type performance, with results showing varying efficacy; for example, Identity Partner A provided a 10% lift on Chrome mobile and a 2.1% lift on Safari mobile. The study obfuscated identity system names to emphasize Prebid’s open-source, inclusive approach, allowing users to freely choose and test different identity systems. This approach enabled a comprehensive analysis of data across different environments, helping to distill directional insights from the aggregated information collected by the analytics systems.

In the continuation of the same case study with a 95/5 split test, the analysis explored the interactions between device types, browsers, and demand partners, revealing how certain demand partners significantly leverage various identity systems. For instance, Demand Partner A demonstrated a strong preference for Identity System A, resulting in an 11% revenue lift on Chrome mobile, and a 6% lift with Identity System B, highlighting the strategic use of identity systems to optimize performance across different settings.

The case study revealed that identity systems on Chrome mobile exhibited the strongest performance, with a top lift of 10%, highlighting a highly responsive environment for specific identity systems and combinations. These results underscore the importance of a testing framework that accommodates fluctuations based on factors like publisher type, target audience, and timing, enabling publishers to continually adapt and optimize their strategies for various device and browser types.

Nic noted that the adoption of the new module is entirely optional for publishers, and that it’s not built into the Prebid core but is available as an add-on that publishers can choose to integrate depending on their needs with specific identity and analytics partners. While providing a framework for publishers to enable this feature, the actual utilization and integration depend on the maintenance and adaptation efforts of their analytics adapters and internal data pipelines, highlighting the need for ongoing updates and compatibility with the latest versions of Prebid.

Q&As

Q1: Is the percentage split determined at an impression, page, or session level?

A: The percentage split for session-based testing is currently being developed as open source, allowing additional features to be incorporated as needed. Each session will maintain a consistent testing schema throughout.

Q2: When this was being designed, were any analytics providers part of the design discussions? Have any analytics providers been proactive about adding support without pressure from publishers to add the features?

A: No, there weren’t any analytics providers participating in the design discussions. However, we encourage contribution from everyone about this.

Q3: What about pub provided ID that could contain several different signals? Will there be a possibility to get stats by sources configured in this module?

A: The feature is not currently built in, but there’s potential for further development and inclusion in the Prebid Asteroid roadmap. If there’s interest, discussions about its integration are welcome.

Q4: Do the tests have a start and end date or duration period, or do you start and end tests manually?

A: Currently, configuration settings within PBJS are manual, similar to other configurations, but there is flexibility for dynamic adjustments depending on the publisher’s capabilities. Whether manually set by smaller publishers or dynamically through APIs by larger organizations with more resources, the system accommodates both approaches, allowing for tailored implementation.

Q5: Can the same infrastructure be extended to the RTD submodules?

A: No plans for this right now, but it came up in discussions. Let us know if you’d like to open up this conversation and help us build this.

Q6: Does ASTEROID simplify testing of any aspects of bids or is specifically tuned to IDs? Does this support data collection on losing bids as well?

A: Specifically tuned to IEDs right now. It doesn’t support data collection on losing bids at this point.

Q7: Have any of the various companies that provide wrappers around Prebid expressed interest in supporting this?

A: Not right now.

Q8: What Prebid version is this planned for and how many months out?

A: The code is currently being built out, we are looking at 9.3 – 9.4. We will send out more details later.

Q9: Have we discovered anything else which could enrich bid requests, which publishers can learn from, then leverage this enrichment for the buy side?

A: Besides identity systems, publishers can enrich bid requests in various ways, such as contextual data or sending taxonomy. While identity systems currently provide a significant boost to bid requests, other opportunities for bid augmentation exist, including audience segmentation and taxonomy addition.

Q10: Has any consideration been given to pushing the ad server GAM key values through a Prebid configuration, similar to how Prebid pushes other key values? Sounds like a publisher would otherwise have to pull from ASTEROID config in a local storage to get the values.

A: We provides examples of how to send metadata to any chosen ad server, which can be captured directly from the bid one event in Prebid, eliminating the need for publishers to retrieve data from storage. This data, including test scenarios and enabled features, is readily available in the bid one event, allowing for easy access and analysis to guide decision-making. Publishers can integrate this metadata into their existing analytics systems or ad servers to optimize their use of identity systems based on collected insights.

Q11: There’s a current project in Prebid server to develop a general purpose rules engine. Are folks that are working on ASTEROID aware of the Prebid server project and any thought on leveraging the rules engine?

A: The concept of a rules engine has been mentioned, and while it hasn’t been directly linked to this effort, its potential integration could enhance test setup and management by allowing rules to be established from external sources like a private server into the PBJS configuration. This approach could make the offering more robust and accessible for more publishers.

Q12: So when you said that traffic was split upon sessions, what do you consider a session?

A: A session is defined by setting session or cookie storage; for instance, cookie storage might be set for an hour, while session storage duration depends on the browser settings.

Conclusion and Participation

You can find the session’s recording here and technical documentation will be available on prebid.org, with announcements to be shared across multiple channels. 

As always, we encouraged community participation and feedback in the development of this project to further enhance the features. Please reach out to us if you are interested!