Your Google Ads Conversions Are Probably Lying to You

The account looks healthy. Conversions are up month on month, cost per conversion is down, the dashboard is green. Then you talk to the sales team and they tell you the leads are worse than they were in March.

Both things are true. The account is performing exactly as instructed — it has just been instructed to buy something nobody wanted. Almost every underperforming Google Ads account we inherit is not badly built. It is optimising sincerely toward the wrong target, with considerable skill.

What "conversions" usually turns out to mean

Open the conversion actions list on an account nobody has audited in a year and you will typically find some combination of the following, all set to "primary", all feeding the bidding algorithm equally:

• The contact form submission. Fine.
• A thank-you page view, firing on the same submission. Now it counts twice.
• Any click on a tel: link, including the accidental ones, including the same person clicking three times.
• A newsletter signup, which is not a sales lead in any meaningful sense.
• A PDF download from a page that has nothing to do with buying.
• Something imported from an old GTM container that nobody can now explain.

Each of those is a signal saying "find me more people who do this". Together they describe a person who fills in forms, downloads things and clicks phone numbers — which is a real behavioural profile, and not the profile of a buyer.

Why smart bidding makes this worse rather than better

Manual bidding is forgiving of bad measurement, because a human is still looking at search terms and applying judgement. Smart bidding removes the human and pursues the stated goal relentlessly. Feed it a conversion set contaminated with newsletter signups and it will find the audience most likely to sign up to newsletters, allocate budget to them, and report a triumphant cost per conversion while doing it.

This is worth being clear about, because it inverts the usual advice. Automated bidding is not a shortcut past weak measurement — it multiplies the consequences of it. The better the algorithm gets, the more expensive a bad target becomes.

Offline conversion import is the step everyone skips

Here is the gap that costs the most money. Google can see that a form was submitted. It cannot see that the lead was unqualified, that it never answered the phone, or that it closed for a substantial sum six weeks later. Unless you tell it.

Offline conversion import closes that loop: you send the platform the outcome once you know it, so bidding optimises toward deals rather than form fills. It requires a GCLID captured at form submission, stored against the record in your CRM, and pushed back when the status changes. That is a genuine implementation task, not a checkbox, and it is the single highest-return piece of work in most accounts we take over.

It also changes the conversation with the client, because the account starts reporting something a finance director recognises.

The order that actually works

The first few weeks of any account we inherit go into unglamorous work, in this sequence:

• Inventory every conversion action. What fires it, when, and whether anyone would call it a sale.
• De-duplicate. One event per real outcome. Thank-you page or form submission, not both.
• Demote the soft ones. Newsletter signups and downloads become secondary — visible in reporting, excluded from bidding.
• Agree what counts as a lead. Not the marketing definition. The one the sales team would recognise.
• Wire offline import. GCLID captured, stored, returned with the outcome.
• Only then touch bidding. And expect a fortnight of instability while the strategy relearns.

Clients sometimes find this frustrating, because weeks one to three produce no visible performance change. It is worth saying plainly at the start: the alternative is spending the same weeks optimising confidently in the wrong direction.

The question that decides everything

Step four looks administrative and is the one that actually matters. Ask five people at a client what counts as a qualified lead and you will often get five answers — marketing says a form fill, sales says a booked meeting, the founder says revenue.

Until that is settled, every downstream decision inherits the ambiguity: which conversions are primary, what the target cost per acquisition should be, whether a campaign is working. Programmes stall on this more often than on anything technical, and it costs nothing to fix except an uncomfortable half-hour.

What this cannot fix

Two honest limits. Offline import only works if the CRM is disciplined enough to record outcomes reliably — if deal stages are updated erratically, you are feeding back noise of a different kind, and that is a sales operations problem before it is a marketing one.

And attribution will never be clean. Privacy changes, consent mode, cross-device journeys and modelled conversions all mean some portion is estimated. The goal is not a perfect number. It is a number pointed at the right outcome, measured consistently enough that its direction can be trusted.

Where to start this week

Open your conversion actions list and count how many are set to primary. If it is more than two, you have found the problem. Then ask your sales lead what they would call a good lead, and see whether anything in that list matches.

That is a twenty-minute exercise and it usually explains a quarter's worth of disappointing performance. If you want someone to do the full pass, our performance marketing team starts every engagement here — and agencies who would rather this ran under their own brand can route it through white label PPC instead.