Insight
Performance Max Spent Money Where Nobody Buys
What 6 months of Google Ads actually bought
The owner switched the adverts off before I was given the account. He knew something was wrong. He didn’t know what, and the dashboard wasn’t telling him, because the dashboard said the campaigns were converting.
So the first job wasn’t to change anything. It was to ask the account where 180 days of money went, and what each pound came back with. I read the account through the API rather than the screens, and counted purchases only, not the wider “conversions” figure the dashboard leads with. Here’s what the account said.
Nearly all the shop’s sales come from the UK and the US. A third of the budget reached people there, and that third paid its way several times over. The other two thirds were spread across 205 countries. One country on its own took a third of the entire budget, for more than half a million clicks and not a single purchase.
The second number: about half of everything spent went on YouTube.
And the third: in the conversions column the bidding worked from, the account counted more than 170 baskets for every purchase.
Three numbers, one cause, and the cause is easiest to see from a long way outside Google.
The nail factory
There’s an old joke about Soviet central planning. The ministry sets a nail factory its target by number of nails, and the factory turns out millions of spindly, useless pins. So the ministry changes the target to tonnes of nails, and the factory makes one monstrous nail that nobody can lift. The story is, in all likelihood, a joke rather than history. It survives because anyone who has ever set a target has watched the same thing happen.
The factory isn’t stupid and it isn’t cheating. It does what it was measured on, as efficiently as it can. The failure belongs to whoever wrote the target, because the target was the only instruction the factory had.
A Performance Max campaign is that factory, with better machinery. You give it a budget, a goal and a few settings, and it decides everything else: which people, which countries, which of Google’s channels, how much to bid. Google’s own description is that it serves across all its inventory, “including YouTube, Display, Search, Discover, Gmail and Maps.” It is very good at hitting the number you give it. It has no idea what you meant.
This shop’s campaign was given two instructions, and both of them were wrong in the same way.
Performance Max location targeting: the blank box
The Performance Max campaign had no location set at all. A campaign with no location shows wherever Google shows adverts, which is most of the world. On top of that, its location option was “Presence or interest”. Google ships that as the default and recommends it, and it reaches people who have shown interest in a place as well as people who are actually in it.
The shop’s other campaign, a workaday search campaign, had the UK and the US set, with “Presence” chosen: only people in those places. So the same account held both answers side by side. One campaign was told where the customers are. The other wasn’t, and went looking for the cheapest clicks it could find, which are rarely where your customers live.
The country figures show what the blank box cost. Every pound that reached the UK came back as nearly £6 in sales, and the US did the same. Everywhere else, the same pound came back as 9p.
Google Ads conversion goals: counting baskets
The second instruction was the goal. In Google Ads you mark each tracked action as primary or secondary. Primary actions are what the bidding works towards; secondary ones are only recorded. In this account, adding to cart and starting checkout were both primary, alongside purchase.
So the campaign was asked to get as many “conversions” as it could for the money, and told that a basket counted. Baskets come cheap. People add things to carts in their thousands, especially through a click from a video, in a country that scarcely ever buys from the shop. Purchases come dear. Given the choice, the machine made pins.
That’s where the 170 baskets for every purchase came from. It’s also why the dashboard looked rosy while the bank account didn’t: the conversions column was full of baskets, because baskets were what the campaign was told to count.
The bit that looked good
There’s one more trap, and it’s the opposite kind. When I sorted every search term that led to a sale, there were 19 out of 967. All but 2 were the shop’s own name, in one spelling or another.
Those clicks looked like the sweetest money in the account. But people typing the shop’s name are already looking for it, and the shop already turns up in the free results for its own name. Some of those sales would have happened anyway. Brand search isn’t worthless, but I’d argue its return is the least proven number in the account, precisely because it’s the one that looks best.
What I change first
None of this needs a cleverer campaign. It needs sharper instructions, and these go in before a penny is spent:
- The purchase is the only goal the bidding chases. Baskets and checkouts are still recorded, so they’re still visible, but they stop counting towards the bidding. And before any money moves, I check the purchase actually fires on a real order. If the tracking can’t see a sale, the bidding is working blind.
- The UK and the US, by presence. Only people who are there. Nobody else.
- Shopping first, from the products that sold. The 6-month read showed which products sold through shopping adverts. That’s the evidence to start from, and it’s why the product feed, the adverts and the shop need one person watching all three.
- No YouTube, and brand search as a test. If brand search runs at all, it runs as an experiment that has to show it brings new sales, not ones the shop would have had anyway.
The budget comes down, too. Over those 6 months the UK and US got about a third of the spend. A budget aimed only at them can be lower than what the shop was paying and still be more than its customers ever saw.
Not a penny has been spent under the new settings yet. I’ll know within a month whether they earn, and the read that decides it will be the same one: where did the money go, and what did it buy.
Read the account before you touch it
If you’re running Google Ads on autopilot, or you’ve taken over an account from someone else, don’t start by adjusting bids or rewriting adverts. Start with three questions the dashboard won’t put in front of you:
- Where did the money go, by country? Not where you think you’re advertising. Where the spend actually landed.
- Where did the money go, by network? Search, Shopping, YouTube, Display. Performance Max will tell you if you ask it.
- What counts as a conversion? Open the goals and see which actions are primary. If a basket counts, the campaign is buying baskets.
Every setting in an ad account is an instruction to a factory that will follow it to the letter. I think the defaults were written for the factory’s owner, not for you. Write your own.
If your Google Ads have been running for a while and the conversions column looks fine, a read of where the money actually went will tell you whether it really is.
Related: The Dashboard Is a Decoy — the other half of this: what an account hides that only a read through the API shows. One Shopify Expert Operator or Three Marketing Agencies? — why the shopping feed, the ads and the site need one person watching all three. Google Consent Mode v2 and Enhanced Conversions — when the bidding can’t see the sales at all.
Paying for Google Ads That Don’t Sell?
We read where your ad spend actually went and what it bought, then run campaigns where every click is tracked and every pound is accounted for. Before we speak, Anthony reads your site against what you want to be found for, so the first call starts with what we found.
Or email tony.cooper@cooperagency.co.uk. Every enquiry is answered within 24 hours on working days.
Anthony Cooper
Founder, Cooper Agency
Anthony Cooper leads every Cooper Agency engagement. If this piece describes a problem your business is facing, we would be glad to talk it through.
Contact