The first thing we check in any account is not the structure, the keywords or the creative. It is what the account is optimising towards.
Why this is the first check
Smart Bidding, Advantage+, Performance Max, App Campaigns – all of them are the same machine underneath. You define a target, the system finds more of it. That is genuinely powerful, and it is also why a wrong target is so expensive: the system will pursue it relentlessly and report success the whole time.
In the accounts we inherit, something in the conversion setup is wrong more often than not. Not exotic wrong. Ordinary wrong, in about five recurring ways.
The five that recur
1. The tag fires on page load
A thank-you page tag placed on a template rather than on the confirmation itself, or an event bound to a button render instead of a click. Every visit becomes a conversion. Cost per conversion looks superb. Bidding piles budget into whatever produces cheap pageviews.
The tell: a conversion rate that is implausibly high and remarkably stable across very different campaigns.
2. The same conversion counted twice
A GTM tag and a hard-coded gtag on the same page. Or a platform integration and a manual event both reporting a purchase. Revenue doubles in the interface and nowhere else, and the bid strategy thinks it has twice the headroom it has.
The tell: platform-reported conversions that are close to exactly two times what your own database says.
3. A micro-conversion set as Primary
Newsletter signup, PDF download, “view contact page”. Useful to know, disastrous to optimise towards. The system will find the audience that downloads things, because that audience is cheap and abundant, and it will stop looking for the one that buys.
The tell: rising conversion volume alongside flat or falling revenue.
4. Value that is not passed, or passed wrong
Ecommerce accounts where every purchase is assigned a static value. Lead-gen accounts running value-based bidding on a value nobody has defined. Subscription apps optimising on trial starts with no signal about which trials convert.
The tell: a Target ROAS strategy in an account where the revenue field in the conversion report is a round number repeated every time.
5. Cross-device and consent gaps counted as reality
Consent mode set to reject-by-default with no modelling, or Enhanced Conversions never enabled, so a large share of real conversions never make it back. The system then underbids on the segments where it is blind, which are frequently the valuable ones.
The tell: a channel your own analytics rates highly that the ad platform insists is performing badly.
How to audit this in an afternoon
Pull the list of conversion actions and, for each one, write down three things: what it is supposed to represent, whether it is Primary or Secondary, and what your own database says the count was last month.
Any row where the two counts disagree by more than about ten percent is a problem. Any Primary row that does not correspond to money is a problem. Any account with more than two Primary rows needs a conversation about which one is actually the target.
Then fix in this order: stop the double counting, demote the micro-conversions, correct the value passing, and only then look at bid strategy settings. Changing the strategy before the signal is right just teaches a new wrong lesson.
What to expect afterwards
The reported numbers usually get worse. Conversions fall because you stopped counting things that were not conversions. Cost per conversion rises for the same reason. Meanwhile revenue holds or improves, because the budget is now going somewhere else.
This is the uncomfortable conversation in the first month of most engagements, and it is worth having once rather than managing around forever.
If you want this done on your account, the free audit starts here every time. Related: what PPC management costs.