Decoding the open internet: Supply path optimization

While many marketers often focus on well-known topics such as creative best practices, generative AI, privacy trends, and app store optimization, there are lesser-known topics of in-app advertising that are equally — or perhaps even more — critical for marketing success.

This is especially true for in-app advertising on the open internet, a vast, decentralized network of millions of independent apps that reach billions of users. With the right strategies, marketers can tap into this expansive, incremental channel to unlock significant growth and take their app to new heights.

This blog series will explore the open internet's mechanics, complexity, and opportunity for app marketers. These topics are often less discussed, but understanding them will help marketers grow their businesses more effectively. This three-part series will cover:

• Supply path optimization (SPO): The open internet is a diverse ecosystem that spans publishers, mediation platforms, supply partners, ad networks, and DSPs. These players are all interconnected, leading to multiple competing bid requests for a single ad impression. SPO cuts through this clutter to identify the most efficient path to serve an ad by predicting a user's potential value, the bid price to win an impression, and serving the highest-performing creative.
  
  
• Moloco software development kit (SDK): Next, we will introduce the Moloco SDK, a direct path to publishers that enables Moloco's powerful machine learning engine to streamline the ad buying process. This unique SDK will drive better returns for marketers and publishers by reducing intermediary fees, enhancing targeting, and enabling greater creative control.
  
  
• Transparency and control of measurement and creative rendering features: Finally, we'll look into the importance of transparency and control of measurement and creatives to ensure marketers make accurate, fair performance comparisons across channels.

Part one will focus on supply path optimization, covering the distinctions between the open internet and walled gardens, a deep dive into open internet ecosystem players, and why advanced machine learning is necessary for marketers' success. 

‍
‍Walled gardens versus open internet

In contrast to the open internet, walled gardens are controlled environments where platforms such as Google and Meta serve ads on their owned and operated inventory. These ecosystems define their auction dynamics and have specific standards for creatives and measurement. On the other hand, in-app advertising on the open internet consists of an interconnected ecosystem with millions of independent apps, mediation platforms, supply partners, ad networks, and DSPs.

While many app marketers may be accustomed to advertising within walled gardens, the open internet offers immense, incremental scale and, when leveraged effectively, presents limitless growth opportunities.

Overview of the open internet ecosystem for mobile advertising 

The open internet ecosystem for mobile advertising is often viewed as a linear path where bid requests flow from the supply side (publishers) to the demand side (ad platforms). These ad platforms then bid to ultimately win the impression to serve an ad within the publisher's app. 

In this linear view, publishers connect to mediation platforms — SDKs that manage ad slot auctions among exchanges — to maximize effective cost-per-thousand impressions (eCPMs) and fill rates. Mediation platforms are connected to exchanges (e.g., ad networks and independent supply partners) that aggregate demand from ad platforms that ultimately serve ads on behalf of marketers.

While the simplified view helps us understand the key players in the ecosystem, it doesn’t reflect its true complexity.

Here’s the reality of the open internet: When a user accesses an app, the available ad inventory is auctioned through a mediation platform connected to multiple exchanges. As a result, a single ad impression can generate numerous bid requests from multiple exchanges, each competing to serve an ad to the same user.

The implications for marketers

Why is the reality of the open internet this way? Various players across the value chain typically have conflicting interests — mediation platforms and exchanges aim to maximize publisher revenue, while marketers focus on achieving the best ROI. These contradictory goals can lead to strategies that, while advantageous for one party, may result in increased costs and complexities for the other. Here are the fundamental dynamics to consider:

• Bid duplication: It's common for exchanges to generate multiple bid requests for the same ad impression to maximize publisher revenue by driving up competition. This is often the result of various bidding methodologies where the same exchange may leverage waterfall and header bidding. While the ecosystem is starting to shift primarily to header bidding, which is an open auction where the highest bidder wins, waterfall bidding is still prevalent, where ad inventory is auctioned sequentially at preset prices that have predetermined bid floors.
  
  For instance, as illustrated in the example above, an average of 12 bid requests occur per ad impression across three or more exchanges. This bid duplication can complicate the bidding process as each exchange intermediary takes a cut and can lead to inflating the price for an impression. Additionally, marketers can bid against themselves across different exchanges, further hurting their ad performance.
  

• First-price auctions: Most bids are first-price (1P) auctions on the open internet. This means that you pay exactly the amount you bid if you win the impression. Winning these auctions requires precisely predicting a user's value to a specific app and accurately modeling the winning bid necessary to display the ad to that user. If the estimated user value is higher than the bid price needed to win the auction, that marketer will likely have a positive ROI. 
  
  Conversely, walled gardens typically utilize second-price auctions, where marketers set a maximum bid for a user or in-app action. In this auction, the winning bid pays slightly more than the second-highest bid. However, the open internet's 1P auctions mean it's critical to precisely identify high-value users and pay just enough to win the impression and maximize ROI.  

• Creative capabilities: While a bid request is for a specific type of creative format, there are differences across exchanges in how a creative is rendered. For example, consider two hypothetical supply paths for a video ad slot: Path A's bid request will go through one exchange, and Path B's will go through a different exchange. Path A may have additional creative rendering features such as a different size X button, end card support, or StoreKit creative features. While for the same type of creative format, these other paths may have very different results for marketers.
  
  This necessitates using advanced machine learning to analyze and choose the most effective supply path, considering the potential ROI driven by the creative rendering capabilities of each available path for a given impression.‍

There is no exclusive inventory

Another important implication for marketers to understand is that there is no exclusive inventory on the open internet.

Therefore, success on the open internet doesn’t come from owning “premium” inventory but from strategically accessing it to choose the best supply path for maximizing ROI. 

Optimizing for the right path on the open internet

With the open internet’s numerous pathways for an impression and its non-exclusive inventory, the ostensibly straightforward goal of bidding the right price with the right creative is…not so simple. This can become increasingly overwhelming for the human brain to handle, which is precisely where advanced machine learning (ML) excels.

Advanced machine learning is at the core of supply path optimization because it can process vast amounts of data, make predictions in milliseconds, and continuously learn and improve over time.

Machine learning’s effectiveness lies in its ability to manage and select the optimal path, making near real-time predictions to maximize ROI. Here's how ML successfully handles this:

• Bid requests: Machine learning determines the value of a specific user and how valuable that user is to a specific app. By comparing these values against predicted win prices, ML strategically places bids to avoid multiple bids for the same impression, which can drive up costs. Ad platforms have around a 100-millisecond window to respond to a bid request to serve an ad, and machine learning autonomously decides the best path in this brief window to maximize ROI and reduce unnecessary costs.

• Pricing: Given pricing on the open internet is largely based on 1P auctions, machine learning utilizes predictive algorithms to analyze first-party, contextual, and campaign data, enabling it to estimate the optimal bid price based on a potential user's value. This helps prevent overbidding and ensures ads are delivered efficiently and cost-effectively.
  
  
• Creatives: Machine learning also predicts the effectiveness of different creative rendering features across supply paths by analyzing real-time and historical data. It identifies which supply path is likely to maximize user engagement and ensures a creative with the optimal creative rendering features is served to a specific user.

Advanced machine learning is the answer to successfully navigating the complexities of the open internet. An effective machine learning system that can analyze immense amounts of data—including first-party, contextual, and campaign data—automatically chooses the optimal supply path to maximize marketers' profitable outcomes.

Supply path optimization key takeaways 

As we've explored in part one of this series on in-app advertising on the open internet, here are key takeaways:

• Open internet is an incremental channel: The open internet offers immense opportunities for marketers by providing a broad and dynamic environment that, while complex, allows for significant reach and engagement beyond the confines of walled gardens.
  
  
• There is no exclusive inventory on the open internet: Inventory on the open internet is not unique — DSPs can access a vast range of inventory across exchanges. In contrast, ad networks often access only the publishers integrated with their SDK. Understanding this helps marketers strategize more effectively and streamline their ad buying process.
  
  
• Machine learning is needed to find the optimal path for success: Supply path optimization is necessary for navigating the open internet, and machine learning is the key to making sense of it all. Machine learning simplifies the open internet’s complexity by choosing the right supply path, bidding the optimal amount based on a user’s predicted value, and serving the ideal creative to maximize user engagement.

Our next topic in this series will explore the upcoming launch of direct integration with publishers through the Moloco SDK. And this SDK is unique to the industry — stay tuned to learn more!

‍Any third-party links are provided for your information only. Such links should not be interpreted as approval by Moloco of those linked websites or any information you may obtain from them. Moloco has no control over the contents of those sites or resources.

‍