Performance Max spends your budget on whatever it believes is the cheapest path to the conversion you asked for. That is the whole mechanic, and almost every complaint about the campaign type comes from not accepting it. It is not looking for new customers unless finding new customers is the cheapest way to hit your target. It is not protecting your margin unless you told it what your margin is. It is solving the problem you set, quickly and literally.
So when the spend goes somewhere you did not want, the useful question is not how to stop the system doing that. It is what in your setup made that the cheapest answer. Nine times in ten the answer is one of three things: the conversion you are optimising for is too easy, the target you set is too generous, or the campaign is allowed to buy demand you already had.
We treat the campaign as a machine with three inputs you actually control. What counts as success, what you are willing to pay for it, and which products and audiences are in the room.
It finds your existing demand first because that is cheapest
People who already know your brand convert at a higher rate and cost less to reach. A system rewarded for cheap conversions will go there first and stay there as long as you let it. That is why a new campaign so often looks brilliant in week one and ordinary by week six. Week one was your own audience being bought back.
This is not the system cheating. It is the system doing what a rational buyer does with an unlimited shopping list and a fixed price target. The problem is that you already had those customers, so the reported return overstates what the campaign added.
The fix is to make the campaign show its work. Look at brand searches and direct sessions before the campaign started and after. If total orders barely moved while campaign-reported orders climbed, you are paying for demand you owned. Where it is available to you, exclude brand terms from the campaign so it has to go and find someone new. Where it is not, run the campaign at a target set from what new customers are actually worth, not from a blended number that your own audience props up.
When the spend goes somewhere you did not want, ask what made that the cheapest answer.
The target sets the price, the budget sets the volume
These two settings are often confused, and confusing them is expensive. Your return target is a price. It says how cheaply the system must buy a conversion before it is allowed to buy one at all. Your budget is a ceiling. It says how much you will let it buy at that price.
Raise the budget with a strict target and often nothing happens, because there is not enough inventory available at that price. Stores read this as the campaign being broken. It is the campaign refusing to overpay, which is what you asked it to do.
Loosen the target and spend appears immediately, because a larger pool of auctions now qualifies. The orders that come with it are cheaper to win but usually worth less, so your reported return falls even though your total profit may rise. That is a trade you should make deliberately.
A worked example makes the trade visible. Say you are at $10,000 spend and a return of 4, so $40,000 in revenue. Loosen to a return of 3 and spend $18,000, and revenue becomes $54,000. The ratio got worse and the business got $14,000 more revenue. Whether that is a good trade depends on your gross margin and your capacity, which is exactly why the target belongs to you and not to the algorithm.
Product groups are the steering wheel
The strongest control you have inside this campaign type is which products it is allowed to sell. Listing groups let you split the catalogue, and splitting the catalogue is how you stop the system pouring the budget into whatever is easiest to sell rather than best to sell.
We usually split three ways. Products with room in the margin get their own campaign and a target set from what you can afford to pay. Products with thin margin get a separate campaign and a stricter target, so they can still sell but never at a price that loses money. Products that do not yet have sales history get a small budget of their own, because otherwise they never get shown and never accumulate the data that would let them compete.
The common mistake is splitting by category because that is how the store thinks. Categories mix a $15 accessory with a $300 centrepiece, and the bidding cannot serve both at one target. Split by economics, then name the campaigns after the economics so nobody quietly merges them again in three months.
Reading the spend pattern in the first month
Give a new campaign a fortnight before judging it, then read the pattern rather than the headline. Four things tell you what is happening underneath.
Daily spend that swings wildly from near zero to double the budget usually means the campaign is data-poor and guessing. Narrow the products or raise the budget so it has a stable base to learn from.
Spend that arrives instantly and fully on day one, with a strong reported return, usually means it found your brand audience. Check total store orders against the previous month before celebrating.
Spend that concentrates on a handful of items while the rest of the catalogue goes dark is normal and often correct, but check whether the dark items are dark because they are unprofitable or because they were never given a chance.
Spend that will not reach the budget at all is a target problem, not a demand problem. Either loosen it in steps, or accept that your product economics do not support the volume you hoped for at this price.
What we change before we blame the campaign type
The order we work in is the same every time, because it goes from cheapest fix to most disruptive.
First, the conversion definition. One conversion action, the purchase, with a value attached that reflects margin rather than revenue. Everything downstream is aimed by this.
Second, the product data. This campaign type sells from your feed, so bad titles and missing attributes limit it exactly the way they limit any Shopping campaign. A campaign cannot show a product that Merchant Center has quietly disapproved.
Third, the split by economics described above, with a target for each group that you can defend from your own numbers.
Fourth, the assets. Text and images matter for the placements outside the shopping results, and a campaign with one thin asset group is choosing on your behalf to be bad at those placements.
Only after all four would we conclude the campaign type is wrong for this store. It sometimes is. But we have never fixed an account by switching campaign types while leaving those four inputs where we found them.