Demystifying the Greenbids Real Time Data Module: Smarter Bidder Filtering for Prebid Server

Modern header bidding relies on optimizing every auction to deliver the best results for publishers. But not every bidder deserves a seat at the table for every impression. Sending bid requests indiscriminately can eat into revenue—through unnecessary costs, increased latency, and wasted resources.

The Greenbids Real Time Data (RTD) module introduces an AI-powered approach to filtering bidders in Prebid Server. By predicting which supply-side platforms (SSPs) are likely to respond competitively, this module helps publishers streamline their auctions, cut noise, and focus attention where it pays off.

How the Greenbids RTD Module Filters Bidders

At the heart of the Greenbids RTD module is predictive filtering. Rather than sending every impression to every SSP, the module leverages a machine learning model to estimate, for each impression, which bidders are likely to participate meaningfully. Only those with a high probability of bidding are included in the auction. This technique is a significant shift from the traditional ‘shotgun’ approach.

Core Mechanism

The module reviews the bidders listed in each OpenRTB bid request and predicts, in real time, the probability that each will bid. Two AI-powered artifacts drive this process: an ONNX-format machine learning predictor and a set of probability thresholds. If a bidder’s probability exceeds the threshold, the request proceeds—otherwise, it’s filtered out at the Prebid Server layer.

Performance Constraints

Publishers can set performance-related constraints—such as required win rates or minimum engagement ratios—ensuring the AI doesn’t become too aggressive or restrictive. These settings let publishers maintain revenue control while still enjoying the efficiency of smarter filtering.

Configuration and Real-World Implementation

Activating the Greenbids RTD module requires precise configuration, both at the account and individual bid request level. Publishers must integrate keys and thresholds to determine the module’s filtering aggressiveness and coverage. It’s not a set-and-forget feature; careful calibration is essential for optimal results.

Account and Publisher-Level Controls

Prebid Server supports granular configuration via YAML files and request extensions. You can prioritize settings per auction or fall back on default account parameters. For instance, a publisher can fine-tune ‘targetTpr’ (retained opportunities) and ‘explorationRate’ (percentage of unfiltered traffic used for AI training) per site or even per ad unit.

Concrete Example: Typical Flow

Suppose a publisher activates the Greenbids RTD in their PBS-Java instance. For each incoming auction, the module fetches its AI models and thresholds from a managed Google Cloud Storage bucket. It then predicts bidder engagement, applies publisher-specified constraints, and filters the list accordingly before bid requests are sent. Results (including which bidders were kept or filtered for each impression) are tagged for later analytics and troubleshooting.

Integrating the RTD Module with Analytics and Reporting

Filtering bidders is powerful, but transparency is essential. The Greenbids RTD module integrates directly with analytics tags, feeding actionable data into analytics pipelines and reporting suites. This feedback loop helps publishers track how filtering decisions impact their revenue, win rates, and partner relationships.

Analytics Tags in Action

When an auction runs, the RTD module populates detailed tags for each impression: a unique fingerprint, a map of bidders retained or blocked, and whether the impression was part of the AI’s training (‘exploration’) pool. These tags flow through analytics adapters to reporting tools, informing both revenue monitoring and troubleshooting.

Value for Ad Ops Teams

Real-world example: After deploying the RTD module, an ad ops manager notices a drop in request volume to certain SSPs—while overall revenue per mille (RPM) holds steady or improves. Because the module’s analytics tags clearly identify filtered-out bidders, the team can isolate root causes and evaluate whether model adjustments are needed.

Best Practices and Pitfalls to Avoid

Deploying AI-driven filtering is not just about flipping a switch. To ensure stability and revenue growth, publishers must actively manage, monitor, and refine their RTD configuration. Many common mistakes are avoidable with the right operational habits.

Practical Recommendations

– Always pilot the module on a subset of traffic using the ‘explorationRate’ parameter for safe model training and validation.
– Monitor bidder filtering rates and yields closely; sudden drops may indicate overly aggressive thresholds or model issues.
– Use analytics tags to troubleshoot discrepancies with partners and to optimize account-level settings.
– Refresh AI models and thresholds on the prescribed schedule to capture changing bidder behavior.

Common Mistakes

– Setting ‘targetTpr’ too low and unintentionally restricting competition.
– Forgetting to enable the Spring Boot property in PBS-Java, resulting in a non-functional setup.
– Not informing SSP partners about changing traffic patterns, risking account-level misunderstandings.
– Overlooking cache expiration settings, which can cause models to drift out of sync with live auction dynamics.

What this means for publishers

Smarter bidder filtering translates to streamlined auctions, lower infrastructure costs, and fewer wasted bid requests. Publishers gain more control over auction health, avoid sending non-competitive traffic to SSPs, and ensure that only relevant partners participate in each impression. This can free up resources for more strategic yield optimization and reduce the risk of latency or timeout issues during peak demand.

Practical takeaway

Implementing the Greenbids RTD module is a concrete step toward making every auction more data-driven and efficient. Start with conservative thresholds and a small exploration rate to validate the model’s behavior before scaling up. Make routine use of analytics tagging to keep operations transparent and support effective partner communication.

Treat bidder filtering as an ongoing optimization process—not a one-time project. Regularly review filtering outcomes, keep your configuration and AI models up to date, and involve your ad ops and analytics teams in continuous tuning. Ultimately, a well-managed RTD setup helps you maximize value from your header bidding stack while keeping full control over auction performance.