An inherited Google Ads account where the conversion data was fiction, a leftover tag was inventing its own results, and a second ad platform had been quietly billing for three years without producing a single verified lead. Nobody had looked.
The account came to me the way most rescue accounts do: already running, already spending, already producing reports that looked fine. Money went out every month. Leads came in. The dashboard showed conversions. On the surface there was nothing to fix.
The reports looked fine because the reports were being generated from numbers that weren't real. That's the part worth sitting with: an account can be comprehensively broken and still produce a tidy monthly summary, because the summary is built from the same corrupted data as everything else. It took opening the account and checking things individually, in a live browser, not in a report, to see it.
Roughly $4,400 spent across three years on a second search platform, producing no verified leads at all. Not poor performance. Nothing. It ran that long because the invoices were regular and unremarkable, and because no report ever put the spend and the results on the same page. It was shut off entirely.
The conversion code was executing before Google's tag had finished loading. When that happens the event evaporates: the lead is real, the customer is real, and the account never hears about it. Compounding it: goals that counted things which weren't leads, and a leftover third-party tag container injecting conversions of its own. The account was simultaneously missing real customers and inventing fake ones.
In one campaign, Google identified the search terms behind only about two thirds of the money. The rest sat in an unnamed privacy bucket. And the campaign's only conversion came from inside that hidden portion, meaning every single click Google was willing to show me had produced nothing at all.
Twenty-six clicks went to people searching for the rubber drip mats that sit on cocktail bars. I have never sold those. Also caught: disposable paper mats for auto service bays, and custom silicone mats. Three product categories the business doesn't carry, all paid for at full price, all invisible unless someone actually reads the search terms report line by line.
Six broad commercial terms consumed 39% of the main campaign's budget and returned nothing. That's where price-shoppers and comparison browsers live: high volume, high cost, no intent to buy from anyone.
Two campaigns were competing for the same searches. One was set to Maximize Clicks, a strategy with no concept of a conversion, so it drifted toward the cheapest inventory and bought the leftovers the other campaign had passed on. The signature was unmistakable: a very high click-through rate paired with a very low conversion rate. Cheap clicks from people who were never going to buy.
Nothing else is worth doing while the account can't tell a customer from a click. Every bid, every budget, every automated decision Google makes downstream is only as good as the data underneath it.
Firing conditions corrected so events wait for the tag to be ready. Conversions restricted to genuine thank-you pages only. Fake goals deleted. The third-party container and its phantom conversions stripped out. Then verified in a live browser: view-source to confirm exactly one correct tag, console to confirm it was defined at the moment the event fired. Not assumed from a report. Reports were the problem.
The second ad platform: shut down entirely. The campaign bidding against its sibling: paused. Six generic keywords with no conversions: paused. An account-level negative keyword list built for the products I don't sell: bar and spill mats, silicone, disposable paper, automotive liners, bath and welcome mats, mouse pads, placemats. Now past 118 terms. Geography narrowed to physical presence in the states actually served, rather than the default that also targets people merely interested in them.
Budget shifted onto the three keywords actually generating quotes, with bids raised deliberately on the two flagged as below first-page cost. One keyword turned out to have no destination URL set at all and had been quietly inheriting the ad group default. Corrected. Match types tightened. Bidding moved to manual, so a human decides what a click is worth rather than an algorithm optimizing against data I'd just finished proving was wrong.
A review date was set in writing, 2 September, along with the specific questions to answer when it arrived. No changes before then. Several things changed at once, which is a deliberate trade of clean attribution for speed, and it means the result has to be read as a whole rather than credited to whichever change I happen to believe in most.
Months had been lost to re-investigating the same two suspicions. Both were settled with evidence and written down as closed, not to be reopened without new proof. An audit that leaves everything perpetually "under review" isn't thorough. It's unfinished.
Closed: tracking works. All three lead paths verified end to end in a live browser on 19 August. Any future claim that tracking is broken needs fresh evidence, not an inference from an old observation.
Closed: the thank-you page is fine. Suspected broken for months because it never appeared in Google's report of converting pages. It turned out that report only lists pages where a conversion was attributed to a paid click, and no paid visitor had happened to land there in the windows examined. The page had been working the whole time.
In this account, the form tool recorded roughly sixteen entries in the audit window. Google reported six conversions. That gap is exactly the kind of thing an agency can sell you an expensive fix for. It wasn't a problem. Seven of the form entries self-reported "Google Search" as their source, and people don't distinguish a paid ad from an organic result, so seven self-reported against six attributed is a clean match. Three more leads came from states outside the paid targeting entirely, meaning they could not have arrived through a paid click.
Google only counts a conversion when it can tie it to a paid click. Organic visitors, direct traffic, referrals, and people who typed the company name in carry no such marker. Your form tool will always show more entries than Google shows conversions, and that is normal. I'm putting that on a public page because the alternative, letting a client believe a normal gap is a crisis, is how a lot of unnecessary work gets sold.
Published before the result, so it couldn't be revised afterward to match whatever happened. On the same total budget: roughly a 50% increase in leads per cycle, with cost per lead falling by around 40%. For the first 14-day checkpoint, that worked out to about three conversions at around $120 each.
Conversions did not improve. My projection was overconfident. I was wrong to put a number that specific on a 14-day window. Before concluding the changes backfired, here is what the sample can and can't tell you.
| Metric | Baseline | Round 1 | Change |
|---|---|---|---|
| Click-through rate | 7.21% | 9.01% | +25% |
| Average cost per click | $2.32 | $2.73 | +18% |
| Clicks per day | 6.65 | 9.14 | +37% |
| Conversions | 5 in 49 days | 1 in 14 days | |
| Cost per conversion | $151.37 | $349.55 | +131% |
At the baseline rate of one conversion every 9.8 days, a 14-day window predicts 1.43 conversions. I got one. Run the math: if nothing at all had changed, there was a 58% chance of seeing one or fewer conversions in this window. And even if the changes had worked exactly as projected, you'd still see one or fewer about 20% of the time.
This data cannot distinguish "the changes helped" from "the changes hurt." One conversion is noise, not a signal. Anyone telling you otherwise is reading tea leaves.
What is real at 128 clicks: click-through rate jumped from 7.21% to 9.01%. That's the six dead keywords being gone, and it's exactly what should have happened. Cost per click rose 18%, which is the bid increases doing what bid increases do.
Pool both periods: 63 days, 454 clicks. Group by product intent instead of by individual keyword, and the pattern is decisive.
| Intent | Share of spend | Clicks | Conv. | Cost per conv. |
|---|---|---|---|---|
| Logo mats | 57% | 231 | 5 | $121.70 |
| Anti-fatigue | 24% | 119 | 1 | $258.89 |
| Generic commercial (paused 19 Aug) | 19% | 89 | 0 | none |
Every conversion this account has produced in nine weeks came from logo mat intent. Anti-fatigue has burned a quarter of the spend for one lead. Generic terms burned a fifth for zero. The individual keywords confirm it:
| Keyword | Clicks | Conv. | Cost per conv. |
|---|---|---|---|
| "customized rubber mats" | 26 | 1 | $62.69 |
| [custom logo mats] | 71 | 2 | $115.33 |
| "custom printed mats" | 127 | 2 | $149.96 |
| "commercial anti fatigue mats" | 83 | 1 | $179.46 |
The bid increase on [custom logo mats] worked. Its status now reads plain "Eligible," it cleared the first-page threshold, and it converted at $96.77 in this window versus $133.88 before.
I set the checkpoint at 14 days to test a seven-week projection. Fourteen days at one conversion per ten days will never produce a readable result. That's a design error, not bad luck.
The $120 figure was far too precise for a window with 1.4 expected conversions. A range would have been honest. A single number was theater.
I only reviewed 10 of the 22 keywords on 19 August. The screenshots were paginated and I never asked for the rest. Two of the twelve I skipped, "custom floor mats with logo" and "company logo mats," are logo-intent terms sitting below first-page bid. The only converting category was being starved, and I didn't see it because I didn't look.
Also found in the twelve: two near-duplicate entrance mat keywords sharing 8 clicks and no conversions, two terms Google rarely serves at all, and "commercial anti fatigue mats" drifting below first-page bid, buying poor placement and returning nothing this period. I left it alone deliberately so the window would read clean. It's now the weakest thing in the account.
The evidence says logo intent is the only category with a signal, and two logo-intent keywords were sitting below first-page bid the whole time. So the change is the one that puts more of the same budget into the category that converts, and nothing else moves.
"custom floor mats with logo" and "company logo mats" raised from $3.00 to $3.75, the same move that worked on "customized rubber mats." Budget held. Anti-fatigue left running. The duplicate left running. One variable, so the next read is clean.
What I expect, written down before the result: at the pooled logo-mat rate, four weeks should produce roughly four conversions. I'm not putting a cost-per-conversion number on it this time. The honest projection is a direction, not a decimal: more of the budget landing on logo terms, and cost per conversion moving back toward the $120s from $349.
The questions to answer on 3 October:
If it misses again, that goes on this page too. A case study that can only ever report success isn't evidence of anything except editing.
Not the keywords view. The actual queries people typed. Most accounts have never had it read line by line, and it is where the bar mats live.
Open the conversion actions table and ask what each one is really measuring. If any of them can fire without a human becoming a lead, every number downstream is wrong.
Individual keywords are noise at small scale. Pool them by what the searcher wanted and the pattern shows up in weeks instead of quarters. In this account it took one table.
Close rate, average order, margin, repeat rate. Without those four numbers you cannot say whether any cost per lead is good or bad, and almost nobody has run them.
Read-only access, thirty seconds to grant, revocable in two clicks. You get a written report with the waste quantified, and if it comes to under $500 a month, you don't pay for it.
Get the auditOr just call: (954) 500-7666