Fix conversion tracking before you touch a bid
There is an order to paid search work, and almost everybody does it backwards.
Bids, budgets, keywords, ad copy — all of it is downstream of one question: does the account know what a good outcome looks like? If the answer is no, every subsequent optimisation moves the account further from where it should be, and smart bidding does so faster than a human would.
The four ways conversion tracking is wrong
It counts the wrong thing. A thank-you page reached by three different forms, only one of which is worth anything. A page_view on a contact page treated as contact. Every one of these teaches the bidding algorithm to buy something you did not want.
It counts the same thing repeatedly. A conversion firing on page load, on a page people refresh or return to. Counting set to "every" rather than "one" for a lead-gen action. The account reports twice the leads that exist, and the reported cost per lead is half the real one.
It counts nothing, silently. A tag removed in a redesign. A consent banner that blocks it before consent, on a site where most people never consent. Conversions do not go to zero — they go down, gradually, which reads like a performance problem.
It counts everything as equal. No values, so a £50 order and a £5,000 order are the same event.
How to check in fifteen minutes
Open the conversion actions list. For each action:
- When did it last record a conversion? An action with none in 30 days is either dead or should be removed.
- Is counting set correctly? "One" for leads, "Every" for sales.
- Is it included in the bidding target? An action not included is not being optimised for, whatever the reports show.
- Does it have a value, and is that value differentiated?
Then compare the total to something outside the ad account — your CRM, your order table, anything. A discrepancy of 10% is normal. A discrepancy of 3x is the finding.
The uncomfortable maths
If conversions are over-counted by 2x, your reported cost per acquisition is half the truth. Every decision made on that number — increase budget, raise target, expand match types — was made on a number that was wrong in the direction that encourages spending more.
That is why this comes first. Not because it is tidy, but because the errors compound in the expensive direction.