Key Points

  • Playwire's machine learning bid shaping algorithm delivered a 12% increase in RPS for publishers over standard bidding: Sites using bid shaping saw a 21% increase in Revenue Per Session in the experimental time period, while sites in the control group saw only a 9% improvement in RPS.
  • The algorithm intelligently reduced bid requests by an average of 17% per session while simultaneously increasing revenue performance.
  • Geographic location proved to be a more significant factor in bid value than device type or browser, with emerging markets seeing the highest bid reduction rates.
  • Individual SSPs demonstrated wildly different bid reduction rates, ranging from 9% to 59%, exposing significant inefficiencies in the programmatic ecosystem.

The Quest for Programmatic Efficiency

The programmatic ad ecosystem has grown into a sprawling, inefficient beast. Every day, billions of bid requests flood SSPs and DSPs, many delivering zero value to publishers while consuming precious server resources and contributing to the internet's carbon footprint.

Most publishers and ad tech providers have simply accepted this inefficiency as "the cost of doing business." They continue sending every possible bid request, regardless of likelihood to convert, because they lack the data and technology to identify which bids actually matter.

At Playwire, we refuse to accept the status quo. Our data science team has developed a machine learning algorithm that identifies and eliminates low-value bid requests while preserving (and even enhancing) publisher revenue. This isn't theoretical – we've tested it rigorously, and the results speak for themselves.

What Is Bid Shaping and Why Does It Matter?

Before diving into our experiment, let's clarify what bid shaping actually is and why it's become such a hot topic in ad tech circles.

 

 

Bid Shaping

Bid shaping represents the strategic filtering of bid requests before they're sent to SSPs and DSPs. Rather than passing every possible impression opportunity to all potential buyers, bid shaping selectively determines which impression opportunities should be sent to which demand partners.

In its simplest form, bid shaping answers a fundamental question: "Is this bid request worth sending?"

This matters for several critical reasons:

  • Server costs are skyrocketing. The exponential growth in bid requests has created massive infrastructure costs throughout the supply chain. Every unnecessary bid request consumes computing resources, bandwidth, and energy.
  • Bid request volume impacts QPS. SSPs only have so many resources to listen to requests. An over-abundance of requests, particularly those that SSP chooses not to bid on, will start to have them listening to less of your requests over time to keep their own costs down.
  • Latency affects user experience. More bid requests mean more network calls, potentially increasing page load times and hurting user experience metrics like Core Web Vitals.
  • Environmental impact is real. The computational resources required to process billions of unnecessary bid requests daily contribute significantly to the digital ecosystem's carbon footprint.

At scale, these inefficiencies compound dramatically. A major publisher might send billions of bid requests monthly, with many delivering zero revenue while consuming substantial resources.

Related: What is QPS? Dig deeper into the concept of QPS and learn more about why SSPs filter requests coming into their pipes.

The Bid Shaping Controversy: Valid Concerns or Fear of Change?

Despite its potential benefits, bid shaping remains controversial in ad tech circles. Publishers and monetization partners have legitimate concerns about implementing any technology that might reduce bid volume.

The primary argument against bid shaping centers on a fundamental fear: what if you filter out bids that would have generated revenue?

Additional concerns include:

  • Potential revenue loss. Critics worry that reducing bid requests might eliminate valuable auction participants, potentially lowering competition and CPMs.
  • Imperfect prediction models. No algorithm can perfectly predict which impressions will generate value, raising fears about false negatives.
  • Lack of transparency. Publishers worry about "black box" solutions making critical revenue decisions without visibility into the decision-making process.
  • Operational complexity. Implementing sophisticated bid shaping requires data science expertise and infrastructure that many publishers lack.

These concerns aren't baseless. Poorly implemented bid shaping absolutely can hurt revenue. An algorithm that aggressively filters bids without understanding the nuanced patterns of buyer behavior will inevitably eliminate valuable opportunities.

This controversy is precisely why we conducted a rigorous experimental test with clear control groups. We needed definitive data to determine whether machine learning bid shaping delivers meaningful publisher benefits or represents a theoretical optimization with limited real-world value.

Deep Dive: Publisher’s Guide to Traffic Shaping and QPS Optimization. Read more for tips on how publishers should be thinking about and approaching traffic shaping.

Inside Our Experimental Design

Testing bid shaping algorithms presents unique challenges that standard A/B testing can't address. The traditional 50/50 traffic split approach falls short because bid shaping's value isn't measured at the individual impression level, but rather in how total volume of bid requests affects an SSPs willingness to bid on the domain on average.

Our experimental design required more sophistication. We implemented our ML bid shaping algorithm across a substantial subset of publisher sites, with control mechanisms to ensure data integrity. A small percentage of traffic still sent all bids to create training opportunities, allowing our algorithm to continuously learn and adjust.

We carefully compared performance between sites using the algorithm and the remainder of our ecosystem maintaining standard bidding patterns. This methodology enabled us to isolate the algorithm's impact while accounting for seasonal and market fluctuations.

The experiment prioritized real-world performance over theoretical optimization. Revenue Per Session served as our north star metric, cutting through the complexity to answer the only question that truly matters to publishers: "Will this make me more money?"

Revenue Impact: The Bottom Line

Publishers care about one thing above all else: revenue performance. Everything else – viewability, fill rate, even CPMs – ultimately matters only insofar as it affects the bottom line.

That's why our primary success metric focused on Revenue Per Session (RPS), which measures the average revenue generated each time a user visits a publisher's site. This holistic metric captures the full impact of our algorithm on publisher economics.

The results were definitive and impressive. During the test period:

  • Control sites (without bid shaping) saw a 9% increase in RPS
  • Sites with ML bid shaping experienced a 21% increase in RPS

This 12-percentage-point advantage represents significant revenue growth for publishers – all achieved without changing content, user experience, or ad layouts. The algorithm simply eliminated wasteful bid requests while maximizing valuable ones.

Bid shaping revenue impact (1)

For a publisher generating $100,000 monthly, this difference translates to an additional $12,000 in monthly revenue – $144,000 annually – without any additional traffic or operational changes. That's the power of algorithmic intelligence applied to the bidding process.

Case Study: Learn how a major Utility & Education Website achieved 168% higher CPMs through Traffic Optimization.

Major Utility & Education Website Case Study

Bid Reduction Patterns: The Algorithm in Action

The true brilliance of our ML algorithm appears in its selective approach to bid reduction. Rather than applying a blunt, one-size-fits-all strategy, it identifies specific patterns and contexts where bid requests deliver minimal value.

Overall Efficiency Gains

Across all test sites, the algorithm reduced bid requests by approximately 17% per session. This reduction represents eliminated waste – bid requests that historically generated little to no revenue but consumed server resources and contributed to ecosystem bloat.

The 17% reduction may seem modest, but consider the scale: in a system processing billions of requests daily, this represents billions of unnecessary server operations eliminated while simultaneously increasing revenue.

SSP-Specific Reductions

Perhaps the most revealing findings came from SSP-specific bid reductions. Our algorithm didn't treat all demand sources equally – it learned which SSPs delivered value in which contexts:

  • Average SSP bid reduction: 19%
  • Highest bid reduction for an individual SSP: 59%
  • Lowest bid reduction for an individual SSP: 9%

Percentage Fewer Bid Requests Made by SSP

These dramatic variations expose significant inefficiencies in how SSPs value inventory. The wide variations in bid reduction by SSP indicates that some SSPs are doing heavy filtering of what they consider to be low value requests, while others are doing very little.

Geographic Intelligence

Percentage Fewer Bid Requests Made by Country

Unsurprisingly, user geography emerged as a major factor in bid request value, with our algorithm applying dramatically different reduction rates across regions:

Country

Bid Reduction

United States

10%

UK

19%

Canada

17%

Germany

21%

India

29%

Brazil

22%

Australia

16%

France

23%

Philippines

33%

Spain

24%

The pattern is clear: countries with higher CPMs (like the US) saw more conservative bid reduction, while emerging markets with typically lower CPMs (like the Philippines and India) experienced more aggressive filtering. The algorithm quickly recognized that in certain geos, a larger percentage of bid requests delivered minimal value and could be eliminated without revenue impact.

This geographic intelligence allows publishers to maximize revenue in high-value markets while reducing unnecessary server load in regions with lower monetization potential.

Device and Browser Patterns

Interestingly, device type and browser showed more modest variation in bid reduction rates:

Percentage Fewer Bid Requests Made by Device

Device Type Bid Reduction:

  • Desktop: 18%
  • Mobile: 18%
  • Smart TV: 16%
  • Tablet: 23%

Percentage Fewer Bid Requests Made by Browser

Browser Bid Reduction:

  • Chrome: 17%
  • Mobile Safari: 17%
  • Edge: 19%
  • Safari: 18%
  • Opera: 21%
  • Firefox: 23%

The relative consistency across devices and major browsers suggests that user context matters more than technical environment when determining bid request value. This contradicts conventional wisdom in ad tech.

Settling the Bid Shaping Debate

Our experiment addresses the core controversy surrounding bid shaping: Does selective bid filtering actually improve publisher economics?

The data provides a clear answer: Yes, when done correctly.

The 12-percentage-point advantage in Revenue Per Session between test and control groups demonstrates that intelligent bid shaping not only preserves revenue but significantly enhances it. This definitively counters the primary concern that filtering bid requests inherently reduces auction competition and hurts publisher yield.

However, the emphasis must be on "intelligent" bid shaping. Our experiment also reveals why simplistic approaches to bid filtering often fail:

  • Geographic nuance is critical. The wide variations in bid reduction rates across countries demonstrate that effective bid shaping must account for geographic differences in monetization patterns.
  • SSP-specific strategy matters. With some SSPs seeing 6x higher reduction rates than others, a one-size-fits-all approach to bid shaping would inevitably damage revenue.
  • Learning is continuous. The allocation of a small percentage of traffic to "learning mode" ensured our algorithm continued to adapt to changing market conditions rather than operating on static assumptions.

This experiment conclusively settles the debate: sophisticated, ML-powered bid shaping delivers meaningful revenue benefits while reducing infrastructure costs and environmental impact. The key lies in leveraging sufficient data and computational intelligence to identify the right patterns, rather than applying blunt filtering rules.

Key Insights for Publishers

Our experiment yielded several actionable insights that challenge traditional programmatic assumptions:

  • Geography trumps technology. The data clearly shows that user location has a far greater impact on bid value than browser or device type. Publishers with significant traffic from emerging markets should be particularly aware that a larger percentage of their bid requests may be delivering minimal value.
  • SSP performance varies dramatically. The wide range in SSP-specific bid reductions (9% to 59%) reveals that not all demand partners value or filter the same kinds of bid requests.
  • Site-specific factors matter. Each publisher site in our test experienced different bid reduction rates based on their unique audience composition and content. This reinforces that one-size-fits-all monetization strategies leave money on the table.

The Future of Intelligent Bid Shaping

This experiment represents just the beginning of what's possible with machine learning in the bidding process. As our algorithm continues to ingest data and identify new patterns, we expect even greater efficiency gains and revenue improvements.

Future iterations will incorporate additional signals and more sophisticated prediction models to further refine which bid requests deliver value. We're also exploring ways to provide publishers with more granular insights into their specific bid reduction patterns.

The programmatic ecosystem has grown increasingly complex, but that complexity contains hidden opportunity. By applying computational intelligence to identify and eliminate waste, publishers can achieve the seemingly impossible: fewer bid requests with higher revenue.

Amplify Your Revenue with Playwire's ML-Powered Bid Shaping

Playwire's Revenue Intelligence® suite now includes this powerful bid shaping algorithm as part of our comprehensive publisher monetization platform. Our technology doesn't just optimize your existing revenue – it fundamentally transforms how your inventory connects with demand.

Publishers working with Playwire gain access to this cutting-edge technology without needing to build their own data science team or machine learning infrastructure. Our platform handles the complexity, delivering the revenue benefits automatically.

Ready to eliminate wasted bid requests while boosting your bottom line? Contact Playwire today to learn how our ML-powered bid shaping can transform your programmatic performance.

Updated Apply Now