Performance Max is brilliant at what it does.
The algorithm learns faster than any team of humans could. It tests thousands of product-audience combinations in real time. It finds the highest-converting keywords, placements, and audiences with a speed and scale that manual optimization can't touch.
But brilliant at what it does isn't the same as brilliant at what you need.
PMax is brilliant at optimizing for one thing: conversion volume per dollar spent. That's Google's goal. Get conversions as cheaply as possible so advertisers spend more budget.
The problem is that your goal might be different. You need revenue per dollar. Or profit per customer. Or strategic market share in a category Google's algorithm has decided is inefficient.
When those goals misalign, you hit the optimization ceiling.
Performance Max Is Brilliant At What It Does
Let me be clear upfront. I'm not saying PMax is broken. The opposite.
PMax does exactly what it's designed to do. It's optimized for conversion efficiency (lowest cost per conversion). It analyzes massive amounts of data. It learns continuously. It reallocates budget toward winning combinations and away from losing ones.
This happens at a scale that's genuinely impressive. In any given day, PMax tests and updates thousands of combinations your team would never think to test. The algorithm finds patterns in your data that humans miss. And it gets faster at finding them.
For some businesses, this is exactly right. If you sell a commodity product, your goal really is conversion volume at the lowest cost. PMax is perfect for that.
But for most retailers, it's only half the picture.
The Data Google Doesn't Have (And Never Will)
What keeps me up at night about PMax is that the algorithm has no idea about your business logic.
When Google's algorithm is optimizing your budget, it sees:
- Product attributes (from your feed)
- Historical performance (conversions, revenue)
- User signals (search history, interests, device)
What it doesn't see:
- Your margin on each product (Google has no access to your cost data)
- Your strategic priorities (what matters for this quarter)
- Customer lifetime value (Google sees one transaction, not future value)
- Seasonal planning (what you need to sell when)
- Brand building (products you run at a loss to build perception)
- Market share goals (products you're trying to grow despite current low performance)
Google could know some of this. They could ask for your margin data or your strategic priorities. But they don't, because that data is confidential and they'd rather keep the system simple.
So the algorithm optimizes with incomplete information.
It sees Product A converting at 8% with €50 CPA and Product B converting at 3% with €100 CPA. It moves budget from B to A. That's efficient optimization.
But it doesn't see that:
- Product B has 40% margin (Product A has 15%)
- Product B is a new market you're trying to build
- Product B customers have 3x lifetime value
To the algorithm, Product B is just inefficient. So it gets cut.
This is the core misalignment. Google optimizes with the data it has. You need optimization with the data you have.
When Brilliant Optimization Goes Wrong
I know a home goods retailer that ran into this hard.
They were selling three categories: Bestsellers, Premium, and Clearance. Most of their revenue came from Bestsellers (70% of volume, 18% margin). Premium was smaller volume (15%) but higher margin (38%). Clearance was their exit strategy for slow-moving inventory (15% of volume, 8% margin).
They let PMax run for 6 months with no business rules. The algorithm did its job beautifully. It found the highest-converting bestsellers and poured budget into them. ROAS was excellent. Conversion costs were down.
But profitability went the wrong way. Here's why:
PMax moved budget from Premium to Bestsellers. The math was simple: Bestsellers converted at 6%, Premium at 2%. Over the course of 6 months, Premium's impression share dropped from 35% to 12%. Revenue from Premium fell 68%.
The bigger issue was what this did to future performance. Premium wasn't just underperforming in that moment; the algorithm was actively hiding it from customers who might have wanted it. Each quarter had fewer Premium conversion data points for the algorithm to learn from, so Premium's position kept eroding. Profitable, but invisible.
The client eventually started tagging premium products as High Priority, signaling to the algorithm that these products were worth investing in despite lower conversion rates. Premium's visibility stabilized. Revenue came back. Total margin improved even though overall ROAS went down slightly.
The algorithm's optimization wasn't wrong. It was just answering a different question than the business needed.
The Optimization Ceiling
This is what I call the optimization ceiling. PMax can optimize really well within a certain range. But there's a limit to what optimization can achieve without strategic context.
Every account hits this ceiling eventually. You can squeeze 5–10% gains out of bid adjustments and audience segmentation. But the bigger moves, shifting budget based on margin, protecting strategic products, growing low-efficiency categories, require you to break out of pure optimization mode and move into strategic allocation mode.
At that point, you have two choices:
- Accept the ceiling and live with suboptimal revenue (you get great ROAS on a smaller set of products)
- Add a control layer that tells the algorithm what you've already decided about your strategy
Most retailers choose option 1 accidentally. They don't know they're hitting a ceiling. They just see good ROAS and assume everything's fine.
Closing the Gap Without Replacing PMax
You don't have to choose between PMax's optimization power and your strategic control. You add a control layer on top.
Define your strategy in business rules: which products are High Priority, which are Standard, which are Low. The control layer writes those signals to your product feed daily, so the algorithm knows what you've decided. Within them, PMax still optimizes at full speed. High Priority products get the signal they deserve, and the algorithm decides which exact products within that segment get shown to which users, at which moment.
This is what we built Expanly to do. Your strategy sets the lanes, the platform enforces them daily, and the algorithm works inside them. Most retailers are still framing this as optimization versus control, and that's the wrong frame. The algorithm was never going to learn your priorities on its own. It was waiting for you to tell it.