Your Ads Account Is Optimised. Towards the Wrong Number.
Why a well-run account can still lose money
Modern bidding is extremely good at what it is told to do, and that is the problem. Give an account a target and it will hit it, by finding the cheapest population that satisfies the definition. If the definition is a form submission, it will find the people most willing to submit a form, who are not the same people as those most willing to buy.
This is why an account can look excellent on every metric in the interface while the sales team reports that the leads are worse than last year. Both observations are true. The interface is reporting faithfully against the goal it was given, and the goal was the mistake.
When an account bids towards form fills, it buys form fills. It will find the cheapest people willing to complete a form, and many of them were never going to buy. The algorithm did its job. The target was wrong.
The gap between a conversion and a sale
In most businesses the conversion recorded on site is a proxy, and its relationship to revenue varies by campaign, keyword and audience. Two campaigns with identical cost per conversion can have entirely different values behind them, and the platform cannot see the difference unless you tell it.
Closing the loop
- Import offline conversions where the sale closes later in a CRM
- Use value-based bidding where order sizes vary meaningfully
- Separate brand from non-brand so the blended number stops flattering the account
- Send back qualified lead status, not just lead created
- Keep consent-aware measurement accurate as cookie coverage falls
What happens next
Reported performance usually gets worse and actual profit usually improves. Volume often drops while revenue holds, because the spend that produced nothing stops. It is worth setting that expectation before the first report, or the change looks like a failure in month one.
Finding out what the account is really maximising
Open the conversion actions list and read which ones are marked primary. That, not your strategy document, is what the bidding is maximising. On most accounts that inherit history from several agencies, the primary set has quietly accumulated a page view, a phone-number click, a newsletter sign-up and a form fill, all weighted equally, which means the system is being told that a newsletter sign-up is worth the same as an enquiry.
Then compare the account’s conversion count with the number of real opportunities your sales team logged in the same period. A factor of three or more between them is routine and it is the whole problem: the platform has been optimising towards a population that mostly does not become revenue, and doing it very competently.
Pointing the account at something real
- Mark exactly one conversion action primary, and make it the closest thing to revenue you can measure reliably.
- Demote everything else to secondary, where it is still reported but no longer steers bidding.
- Send qualified-lead and closed-won stages back into the account from your CRM, with values.
- Give the system two to three weeks to relearn before judging it, because the first week after a signal change always looks worse.
- Re-check cost per qualified lead rather than cost per conversion, and expect the headline number to get uglier while the business gets better.
Judging the change on the old metric
After the signal is fixed, cost per conversion usually rises, sometimes sharply, because the account has stopped buying the cheap conversions that were never worth anything. Teams that judge the change on that number reverse it within a fortnight and conclude the old setup was better. The number to watch is cost per qualified lead, and it should be moving the other way.
Common questions
What if our CRM cannot export conversions back?
Then start with the best proxy you can measure on the site, such as a form that only a serious enquiry would complete, and weight it above everything else. A rough proxy pointed in the right direction beats a precise measure of the wrong thing.
Should we use value-based bidding?
Where you have values worth sending, yes. Where the values are invented to satisfy the interface, no: a made-up number trains the system to find more of whatever you guessed was valuable.
How long before the account recovers?
Two to three weeks for bidding to stabilise, and a full sales cycle before the revenue effect is readable. Plan the review date accordingly rather than checking daily.
Does this apply to Performance Max too?
More so. The less granular control you have over placement, the more the conversion signal is the only steering available.
