The work
Two accounts, and what was actually wrong with them.
Both real, both anonymised. Every figure here is reproducible from the account it came from, which is the only reason it is here.
Case one · storm-response services · closed
We were contracted to make paid search work. We decided it couldn’t, and said so.
They expanded the engagement anyway.
$30–60
what clicks were clearing at, against a $200 cost-per-lead threshold
$1,000+
cost per conversion on one campaign
~$3,000
of committed budget protected by stopping early
What Happened
A severe storm hit the service area and a storm-response business went to market on the emergency demand. The campaigns went live four days after that demand had already peaked. We got access the same evening and changed nothing structural for three days, so the peak and the slide both happened on a setup that was not ours.
Results were near zero and the assumption was a tracking bug. There is always pressure to find one, because it is the answer that keeps the spend running. We checked properly. The tracking was sound. The channel was wrong.
How We Proved It
Four independent reads, plus a fifth against a system that had no reason to agree.
- Change history. Our access began the evening of launch. Our first structural change came three days later.
- An untouched control. Keywords we never edited peaked and fell away before we touched anything.
- External demand data. Search interest spiked with the storm and decayed to roughly a third within two weeks.
- A second account. Three campaigns, two landing pages, one identical curve.
- A second system. The CRM and the ad platform disagreed by a single lead over thirty days.
The Arithmetic Nobody Had Run
Clicks were clearing at $30 to $60 against a $200 cost-per-lead threshold, and one campaign cost over $1,000 per conversion. No amount of optimisation closes a gap that size. That check takes ten minutes, and it is now the first thing we run on any account.
We also found the consent checkbox on the lead form rendering white on white. Invisible, unsubmittable, and nothing anywhere reported an error.
What We Recommended
Stop. Pause the campaigns and move the budget to the channel already producing 80+ leads against paid search’s handful. They accepted, then brought us into a second business.
Case two · heavy equipment rental · active
Five ways a lead could reach them.
Four of which were broken.
Roughly a year of spend and two agencies before us, and the measurement had been dead the whole time.
Conversion actions as found · Jun 1 – Jul 31 2026
| Measurement path | Recorded | Why |
|---|---|---|
| Lead form, thank-you page | 0 | Watching a page on a subdomain that no longer existed |
| Second form action | 0 | Wired correctly. Never fired once, because the form sits in a cross-origin iframe |
| Website phone | 0 | Watching a phone number the site had stopped publishing |
| Duplicate call action | 0 | Redundant, and never fired |
| Calls from ads | 1 | The only working path |
One real lead conversion against $4,000+ of spend, and it was a call, not a form. 90% of impressions were mobile, which changes what “just fix the form” actually means.
What Everybody Believed
Five suspected problems were handed to us as fact. Three did not exist. Networks and geo targeting were already correct, and the campaign being judged on a year of failure had been switched on six weeks earlier. The year of spend belonged to campaigns that were already paused.
The Root Cause
The site had migrated and the measurement layer stayed behind. Conversion actions pointed at a retired subdomain and a replaced phone number. Invisible from inside the ad platform, because on the surface the actions look perfectly configured.
We rebuilt the dead action. It still did not fire, and that was the real finding: the form renders inside a cross-origin iframe we did not build, and a conversion call from inside one cannot reach the tag on the parent page.
Bidding had also been switched to automated by an applied recommendation rather than a decision, optimising toward a goal set in which every action was broken. Not toward the wrong thing. Toward nothing.
You cannot see that from inside the reporting tool. You can only see it from outside.
What We Did
- Turned off automated bidding first. You do not hand an algorithm a target it cannot see.
- Rebuilt the keyword set and negative list, and the campaign cleared its Limited status.
- Rebuilt lead capture across all three sub-accounts, verified on the live forms.
- Shipped a script carrying the tracking across the iframe boundary, then verified it end to end: ad click, landing page, script, iframe, populated CRM field.
Where It Stands
The account is measuring properly for the first time. Every lead now arrives carrying the click that produced it, which is the thing the last year of spend never had. The engagement is ongoing, and next is closing the loop back to the platform so bidding optimises on revenue.
The diagnostic
Four checks.
A written answer.
No call.
Free, and you are not on the hook for anything afterwards. If there is nothing worth fixing, the write-up says so.
Check 01
Your site, on the phone your buyer is actually holding. Whether a lead can finish what they started.
Check 02
Your Google Business Profile. Categories, hours, photos, and the reviews nobody answered.
Check 03
Your tracking, read from your own page source. Whether the ads you pay for report anything at all.
Check 04
Your main service and your city, typed into Google. Whether you appear.