The Signals Performance Max Optimises That You Never See

Google describes Performance Max as optimising across channels using machine learning and real-time signals. But the signals it optimises against, and the trade-offs it makes, are largely invisible to advertisers.

Understanding what you can't see is essential to understanding what PMAX is actually doing.

What PMAX Knows That You Don't

User-Level Data

Google has extensive data on individual users:

• Search history and patterns
• YouTube viewing behaviour
• Gmail content and correspondence
• App usage across Android
• Location history and patterns
• Purchase history through Google Pay
• Chrome browsing data

This data informs who sees your ads. But you don't see which signals triggered which impressions.

Conversion Probability Scoring

For each potential impression, Google calculates a conversion probability. This score determines whether to bid, how much to bid, and which asset to show.

You see the outcome. You don't see the calculation.

Channel Selection Logic

PMAX chooses whether to show your ad on Search, Shopping, Display, YouTube, or Discovery. The selection logic considers:

• Relative cost across channels
• Predicted conversion probability per channel
• Competition levels at that moment
• Your remaining budget and targets

You get a blended result. You don't get the decision tree.

Asset Combination Testing

PMAX tests combinations of headlines, descriptions, and images. It learns which combinations work best for which audiences. But the learning isn't shared in actionable detail—you get "best" and "good" ratings, not the underlying logic.

The Trade-Offs You Can't Evaluate

Volume vs Quality

PMAX may pursue conversion volume at the expense of conversion quality. A target ROAS of 4x can be achieved through:

• 100 high-quality conversions at 5x ROAS
• 200 low-quality conversions at 3x ROAS, averaged up by brand traffic

Both hit target. One is better for your business.

Short-Term vs Long-Term

The algorithm optimises for the conversion window you set (typically 7-30 days). It doesn't optimise for customer lifetime value, repeat purchase probability, or long-term brand building.

Easy vs Strategic

PMAX finds easy conversions first. These may be branded searches, repeat customers, or comparison shoppers who would have converted anyway. Harder but more valuable conversions—genuinely new customers—may be deprioritised.

Cheap vs Profitable

Lower-margin products may convert more cheaply than higher-margin products. PMAX will favour the cheaper conversions, even if your profit is higher on the expensive ones.

What This Means for Management

You cannot optimise what you cannot see. This creates a fundamental tension:

1. Trust the algorithm: Accept that Google knows more than you do and let it run
2. Demand visibility: Use every available signal to infer what's happening
3. Constrain the algorithm: Limit its options through structure, exclusions, and feed control

Most sophisticated accounts do all three—trusting where trust is warranted, investigating where visibility exists, and constraining where control is needed.

Practical Steps Toward Visibility

• Use audience signals: Tell PMAX who to target rather than letting it learn from scratch
• Segment asset groups: Create groups by margin, category, or customer type
• Review Insights tab regularly: It's limited, but it's what's available
• Track new vs returning customers separately: Use your own data, not Google's
• Monitor search term reports: Even limited, they show direction

The goal isn't full visibility—that's impossible. The goal is enough visibility to make informed decisions.

We help brands understand what their PMAX campaigns are actually doing. Request an audit or learn about how we approach PMAX analysis.