In a previous ecommerce role, I spent months trying to figure out why our PMax spend kept climbing while profit barely moved. The reports looked clean and ROAS held steady, but the bottom line wasn't keeping pace. The answer was structural. Google's algorithm was making decisions about our budget without any of the data that actually defined our business.
Google and Meta know everything about user behavior. They see clicks, impressions, add-to-cart rates, dwell time, and repeat purchases. What they optimize for is their own scale, not your business outcome.
What they don't see is the data that decides whether you're making money.
What Platforms Actually Know About Your Business
PMax and Advantage+ optimize against one number: return on ad spend. A pixel fires, someone buys, and the platform logs €150 of revenue against €50 of cost. ROAS 3.0. The dashboard turns green.
That 3.0 only describes the slice of reality the platform can see.
The platform knows your click-through rate, your conversion rate, your revenue per impression, your customer acquisition cost, and your repeat purchase rate. Everything that happens between an impression and a purchase event is visible.
Everything that happens around that purchase is not. What you paid for the product. What it costs to pack and ship. What percentage of buyers return it. How long it sits in inventory. Whether it cannibalizes higher-margin alternatives. Whether you wanted to push it this week or pull back.
This isn't a flaw in the algorithm. The algorithm is doing exactly what it was designed to do with the data it has. The problem is that the data it has is half the picture, and the missing half is the half that decides profitability.
Margins: The Biggest Blind Spot
A setup I see at almost every retailer.
You sell two products. The first lists for €100 and earns you €5 in margin per sale. The second lists for €80 and earns you €32. Same category, same audience, same campaign.
Over the past month, PMax has allocated 70% of your budget to the first product because €100 transactions look bigger than €80 transactions. That's what the algorithm sees, and that's what it optimizes for.
On the first product, you make €5 in margin and spend €15-20 to acquire each sale. Your real margin after ad cost is negative. You're paying to lose money on every transaction.
On the second product, you make €32 in margin and spend €20-25 to acquire the sale. You're profitable with room to scale.
If PMax could see margin, it would flip the budget allocation overnight. It can't, so you keep feeding a machine that's optimizing on the wrong number while your profitability erodes one transaction at a time.
This is the gap Expanly was built to close. When we set up margin-aware rules for Reima's North American operation, their profitability improved 32% in three weeks. Same spend, same traffic, same campaign structure. Different signals reaching the algorithm.
ProfitMetrics and similar POAS tools have built real businesses around this problem and they work better than ROAS-only optimization. But margin is only one of the signals retail profitability actually depends on, and a tool that sees only margin still misses the rest of the picture.
Inventory and Stock Cover
Someone in Stockholm searches "Air Max Size 10" on Google. Your ad fires, the click lands, and the size 10 is sold out. You paid for the impression, you paid for the click, and the customer is now buying from your competitor.
PMax doesn't know your size 10 is gone. It's optimizing to put your Air Max ad in front of every relevant search, regardless of which variants are actually available behind the link.
This is one of the most fixable problems in retail advertising, and it's invisible to the platforms by design.
Stock cover is the calculation that closes the loop. Units in stock divided by average daily sales. If you sell 10 a day and you have 30 in inventory, you have three days of supply. On a high-margin product, that means push harder while the stock is there. On a low-margin product, it means pause and let replenishment catch up.
Fashion retailers face this in a sharper form. Broken size runs are normal in the back half of a season: XS and XXL still in stock, M and L gone. The algorithm sees "Air Max" and drives every size-related search to your site. The buyer who wanted L bounces, and your cost per acquired sale on that product just doubled. Nothing in your dashboard tells you why.
Scandinavian Outdoor built rules on the same signal from the other direction. Slow-moving items with high stock cover got pushed during promotional windows, while full-price products stayed at Standard priority so they weren't crowded out by discount-driven conversions. Sales lifted 32% and cost per sale dropped at the same time. (Full case study)
Returns and True Profitability
This is the one that quietly costs retailers 5-15% of revenue every year without showing up anywhere obvious.
Fashion returns sit at 30-40%. Electronics are closer to 5-8%. Furniture, beauty, and sportswear each have their own profile. If your catalog spans more than one category, your blended ROAS is averaging economics that have nothing to do with each other.
Take two products with the same headline numbers. Both report ROAS 4.0 in Google Ads. The first has a 38% return rate, so the real ROAS after returns is closer to 2.5. The second has a 3% return rate, and its real ROAS is 3.9. The first product is roughly 35% less profitable per acquired customer, and the algorithm treats the two as identical.
PMax doesn't have access to return data. Advantage+ doesn't either. They're optimizing on gross revenue, not on the revenue that actually stays sold.
Once you start adjusting priority logic for category-level return rates, the same pattern shows up every time. Budget shifts away from products with strong-looking ROAS but a high return drag, and toward products whose revenue actually sticks. The biggest swings come from mixed catalogs where high-return and low-return categories were getting averaged into the same ROAS view.
Strategy and Seasonality
The platforms know nothing about your calendar.
You're running a winter clearance. The goal isn't maximum revenue next month, it's emptying the warehouse by March 15 to make room for the spring range. PMax is optimizing for steady-state efficiency, which means pacing those clearance items conservatively to protect margin. You actually need velocity. You need that inventory gone before the new stock lands.
Or you're a Nordic outdoor brand. Q4 is your money quarter. December will run 3-4x the volume of other months, and you want to scale customer acquisition in November so the new customers are primed when gifting season hits. You also know returns will spike in January, and you want to weight your November acquisition spend toward categories with lower return rates.
The algorithm doesn't know about March 15. It doesn't know December is 3-4x. It optimizes on the data in front of it today.
This is why every serious retail team manually overrides PMax somewhere. They pause campaigns mid-month, shift budget between products, and exclude SKUs that shouldn't be running. They're adding their strategy on top of the platform's mechanics, every day, by hand.
Bridging the Gap Without Sharing Sensitive Data
The trust question comes up early in every conversation. You don't share cost data with Google. You don't share supplier prices with Meta. Your business data stays on your infrastructure, and only a priority signal reaches the platforms: High, Standard, or Low. They know which products you want to push this week. They don't know why.
This is the architecture Expanly is built around. Your margin file, your inventory feed, your return rates, and your strategic calendar all live on Expanly. The output is a single value written to a custom label slot in your Google Merchant Center feed. Five label slots are available, and you decide what each one signals.
Setup takes around 30 minutes. You connect your product feed, plug in cost and inventory data, define the priority rules, and the labels start flowing the next day. By the second week, the budget allocation has shifted to match your priorities. Meta support is on the roadmap and the same pattern applies there: your data stays with you, only the signal moves.
Swappie is a good example of what this looks like when several signals stack. Their refurbished phone catalog covers thousands of unique products: different models, conditions, colors, and storage, with margins that swing from 18% to over 40% and supply often capped at a handful of units per configuration. They built a four-tier priority system that combines margin, inventory cover, and supply constraints into one daily label per product. The cost data stays with them. The signal reaches Google. Efficiency improved 20% in the first month on the same budget. (Full case study)
The algorithm is good at what it does. It's just been deciding alone. Once it has the signals that define your business, the same machinery that was eroding profitability starts protecting it.
