Estimated Action Rates and User Value Scores Let Relevant Ads Defeat Higher Cash Bids in Meta's Ad Auction

Every time a user opens a social media feed, a silent, high-speed calculation decides what appears on their screen. The ad auction executes in milliseconds, evaluating thousands of eligible advertisements for a single slot. It determines not just which brand gets the impression, but exactly how much they will pay for it.[3][5]

Unlike a traditional auction where the highest cash bid automatically wins, modern digital advertising systems are designed to balance advertiser revenue with user retention. If a platform simply awarded inventory to the highest bidder, users would be flooded with irrelevant, spammy content. The algorithm intervenes to prevent this outcome.[4][5]

To maintain that balance, Meta does not rank ads by the dollar amount offered. Instead, the system calculates a metric called Total Value for every candidate ad. The advertisement that achieves the highest Total Value score wins the placement, regardless of whether its underlying cash bid was the largest.[3][5]

The Total Value Equation

The mechanism that governs this decision is a specific mathematical formula. The platform calculates Total Value by multiplying the advertiser’s cash bid by an Estimated Action Rate, and then adding a User Value score. This equation fundamentally alters the economics of digital media buying.[5]

Because the Estimated Action Rate acts as a direct multiplier, it disproportionately influences the final score. An advertiser bidding two dollars with a high predicted engagement rate can easily defeat a competitor bidding five dollars with a low predicted rate. The math forces advertisers to prioritize relevance over raw budget.[2][5]

This structure explains why brands often struggle to scale their campaigns simply by increasing their daily spend. If the underlying creative does not resonate with the audience, the multiplier drops. The advertiser is then forced to bid exponentially higher cash amounts just to maintain the same Total Value score.[2][5]

The system employs a Vickrey-Clarke-Groves (VCG) second-price auction model to finalize the cost. Under this framework, the winning advertiser does not actually pay their maximum bid. Instead, they are charged the exact minimum amount required to beat the Total Value score of the second-place advertisement.[1][5]

Predicting the Action Rate

The Estimated Action Rate represents the machine-learning prediction of how likely a specific user is to take the advertiser’s desired action. If a campaign is optimized for purchases, the algorithm evaluates the probability that the user will actually buy the product after clicking.[2][5]

To generate this prediction, the system analyzes thousands of historical data points in real time. It reviews the user’s past behavior, the historical performance of the advertiser’s account, and the specific creative elements of the ad. These signals combine to form a highly personalized engagement probability.[3][5]

A low Estimated Action Rate effectively prices an ad out of the market. If the models determine that a user is unlikely to convert, the multiplier shrinks toward zero. The advertiser’s cash bid is neutralized, preventing the ad from winning the auction even if the brand is willing to overpay.[2][5]

Advertisers can improve their predicted rates by aligning their creative with the specific audience segment most likely to respond. Providing the algorithm with high-quality conversion data through server-side tracking also strengthens the prediction model. When the system trusts the data, it assigns a higher probability multiplier.[4][5]

The User Value Penalty

While the Estimated Action Rate predicts positive outcomes, the User Value component acts as a quality control mechanism. This additive score measures the overall experience the advertisement delivers. It evaluates both the ad itself and the post-click environment on the advertiser’s landing page.[5]

The platform actively penalizes advertisements that degrade the user experience. According to industry analysis, poor creative can reduce the Estimated Action Rate by up to 50 percent and drop the User Value score by 10 percent. Engagement baiting and sensationalized language trigger these automated quality downgrades.[5]

The penalty extends beyond the application interface. If a user clicks an ad but immediately bounces from a slow-loading or irrelevant landing page, the algorithm registers a negative signal. This poor post-click experience reduces the User Value score for future auctions.[4][5]

A severely negative User Value score can completely offset a high cash bid. The platform will throttle delivery for low-quality ads, forcing the advertiser to pay significantly higher costs per thousand impressions (CPM) to achieve any visibility at all.[3][5]

The Shift Toward Creative Relevance

The mechanics of this auction have recently undergone massive architectural upgrades across the industry. Beginning in 2024 and expanding through 2026, platforms are deploying new ad-ranking and retrieval systems that utilize advanced embedding models to process inventory at an unprecedented scale, evaluating up to 10,000 times more ad candidates per session.[3][5]

These upgrades fundamentally shift the weight of the auction toward creative relevance. The systems use deep neural networks to match ad creative to user context, evaluating tens of millions of candidates instantly. Creative quality now outweighs historical account performance in determining the winner.[5]

For performance marketers, this architectural change means that traditional audience segmentation tactics are no longer effective. The algorithm now finds the audience based on the creative itself. Brands are consolidating their campaigns into broader structures, allowing the machine-learning models to handle the targeting.[4][5]

The testing cycle for new advertisements has also lengthened under these advanced systems. Because the models require more data to build accurate embeddings, evaluation windows have extended from 48 hours to between 7 and 10 days. Advertisers must allow the algorithm time to stabilize its predictions.[5]

Strategic Bidding in a Value Auction

Despite the heavy emphasis on relevance, the cash bid remains a critical lever for advertisers. Platforms offer several automated bidding strategies, including volume maximization and value optimization. These settings dictate how aggressively the algorithm pursues the available inventory within the allocated budget.[3][5]

A volume-focused strategy instructs the system to secure as many conversions as possible, allowing the cost per acquisition to fluctuate. Conversely, cost cap strategies prevent the algorithm from bidding on auctions where the predicted cost exceeds the advertiser’s profitability threshold.[4][5]

When an advertiser applies a strict cost cap, they are essentially refusing to participate in expensive auctions. If their Estimated Action Rate is low, the system will simply stop delivering the ad rather than overpaying. This protects the budget but can result in stalled campaigns.[5]

Ultimately, the modern ad auction operates as a self-correcting ecosystem. By linking delivery costs to user experience, the platform forces advertisers to create better content. Brands that invest in high-quality creative and seamless landing pages are rewarded with cheaper inventory and higher reach.[4][5]

Those who attempt to brute-force their way into the feed with massive budgets and poor creative find themselves mathematically blocked. The Total Value equation ensures that relevance remains the dominant currency. In this marketplace, the algorithm’s prediction of human behavior outranks the advertiser’s wallet.[5]

As machine learning models grow more sophisticated, the gap between high-quality and low-quality advertisers will continue to widen. The infrastructure is designed to protect the user’s attention at all costs. Advertisers who align with that goal will win the auction; those who fight it will pay the penalty.[3][5]