Where Google Ads budget actually leaks

The call usually starts the same way. Spend is up, inquiries are flat, and the owner wants to know who set the bids wrong.

Bids are rarely where the money went. In the accounts we audit the bidding is competent, and increasingly it is automated, which means it is doing precisely what it was told to do. The money leaks out of decisions taken when the account was built, and it keeps leaking no matter how many hours anyone spends stacking bid adjustments and ad schedules on top.

Two kinds of waste

Tactical waste is a weak headline, a bad bid, a day of the week that never pays. It costs a few percent, it shows up in the data, and it largely corrects itself once somebody looks.

Structural waste is the kind the system reads as success. A campaign optimising towards a conversion action that fires on newsletter signups will become very good at buying newsletter signups. Nothing in the interface objects. The graphs climb. The bank balance does not. That is what makes structural waste durable: it has the shape of performance, so it survives audits, agency handovers and years of monthly reporting.

The leaks you can find yourself

Five of these turn up repeatedly, and all five are visible from the standard interface in an afternoon.

Geography that does not match where you can actually work. Google’s default for Search campaigns is still “Presence or interest”, which reaches people in your targeted locations and also people who have shown interest in them. For a brand selling across Europe that is reasonable. For a firm that fits kitchens in Varna it means paying for clicks from somebody in Manchester reading about Varna. The alternative setting is “Presence”, and it is one radio button per campaign.

Brand searches blended with everything else. People searching your company name have already decided. Their clicks are cheap and they convert, and mixed into the same campaign as cold traffic they make acquisition look like it is working. Filter the search terms report for your own name, subtract that spend and those conversions from the totals, and look at what is left. That remainder is your acquisition performance.

One campaign carrying work with different margins. Automated bidding optimises for conversion count, or for a blended return. Given the choice it will buy fifty low-margin jobs rather than ten profitable ones and report a strong month either way.

Campaign types with nothing separating them. If a Performance Max campaign and a Search campaign both sit in the account, they are candidates for the same queries. Google’s documentation is explicit about the order: a Search campaign whose keyword exactly matches, or is spell-corrected to, the query is preferred over Performance Max, and where no exact-match keyword applies, the campaign with the higher Ad Rank serves. Which is a tidy rule that creates an untidy problem, and we will come back to it.

Landing pages that were never built for the query. Sending every keyword to the homepage is a structural decision, and it puts a ceiling on conversion rate that no amount of bidding will lift. We wrote separately about where paid traffic dies.

The one underneath all of them

Every leak above is findable once you look. The reason we open the conversions table before touching any of them is that the conversions table decides what looking means. It is the dictionary the rest of the account is written in.

So read every conversion action in it, one at a time, and ask whether you would be happy to pay money for exactly that event. Page views. PDF downloads. Form loads rather than form submissions. Phone number clicks from people who never waited for an answer. If an action would not survive that question, it should not be feeding bidding.

Here is where most advice stops, and where it is wrong. The usual fix is to mark the junk “secondary” and assume it is now cosmetic. Google’s own page on the setting does say secondary actions are for observation only and report into “All conversions” rather than “Conversions”. But it also says that conversion actions inside a custom goal are used for bidding regardless of whether they are marked primary or secondary, and that primary actions you have excluded from optimisation may still be used to enhance predictions. So the label is not a switch. If somebody built a custom goal around a soft action two years ago, demoting that action changes the column it appears in and nothing else. You have to go and look at the goals, not just the actions.

Our piece on conversion tracking that counts real inquiries goes through the setup properly.

What tactical work still earns its keep

None of this makes day-to-day management pointless. Negative keyword maintenance pays for itself repeatedly, asset testing in responsive search ads is real work, and somebody has to watch the query drift that broad match and AI matching produce over time. Our articles on maintaining negatives and reading the search terms report cover both, and ongoing optimisation is most of what we do day to day.

The argument is about sequence. Tactical work layered over a broken structure compounds the error, because you get steadily better at buying the wrong thing.

It is also worth saying what is not structural. Click costs rising in a competitive market are not a flaw in your account. Neither is a genuinely difficult product with a six-month sales cycle. Do not rebuild an account because the auction got more expensive. Rebuild it when it is measuring the wrong event, or mixing things that should have been kept apart.

The question we cannot answer

Back to Performance Max and Search competing inside the same account. The rule Google publishes tells you which campaign wins an impression. It does not tell you what would have happened if the other one had won it.

That is the part nobody can see from inside an account. When Performance Max takes a query on Ad Rank because no exact-match keyword covered it, there is no column that reports whether a Search campaign would have taken the same click for less, or whether the sale would have arrived anyway through organic results or a returning customer. The counterfactual is not in the data, and the experiment that would settle it, switching Performance Max off in a live account for long enough to read a clean result, costs the advertiser real money and still gets contaminated by seasonality.

We do not know how much Performance Max spend in a mature search account is incremental. We have not found a measurement approach that survives scrutiny, and the vendors who publish confident percentages on this are, as far as we can tell, measuring their own clients rather than the question. It is the largest open question in the structure of a modern Google Ads account, and we would rather leave it open here than hand you a number we cannot stand behind.