318 conversions. Roughly $10 a lead. Sixteen months of what looked like the best-performing account we had seen all year.
Twelve real leads.
The gap between those two numbers is not a reporting error. It is a broken Google Ads conversion tracking setup. For sixteen months it taught the bidding algorithm to buy the wrong traffic. The client never knew, because every dashboard they looked at said things were fine. Below is exactly what we found, and the ten-minute check that tells you whether your own account is doing the same thing.
What the dashboard said
A roofing business in Australia. The account had run for well over a year with a previous agency. The monthly reports were good reading: conversion volume climbing, cost per lead holding near $10, plenty of green arrows.
Nobody was lying. The numbers in those reports were the numbers Google Ads was reporting. That is the part that makes this dangerous rather than merely sloppy. There is no villain to catch, and no single moment where it visibly went wrong. The account optimised itself into a hole while everyone involved watched a chart go up. By the time we saw it, the pattern was sixteen months deep and had become the account’s normal.
Following the signal
The first thing we do on any takeover is open every conversion action and ask what it physically requires a human to do. Not what it is named. What it requires.
96% of the 318 were taps on the phone-number button. Not connected calls. Taps.
A tap fires the moment a thumb touches the button. It fires if the person hangs up before it rings. It fires if they misclick while scrolling. It fires twice if they tap, get distracted, and tap again. It measures an intention to maybe call. That is not nothing, but it is a very long way from a lead. It should never have been the signal the account was optimising toward.
Real traceable leads, meaning form fills plus call-extension calls that actually connected, came to 12 over sixteen months. True cost per acquisition was closer to $263.
Why broken Google Ads conversion tracking is worse than a bad report
If the only consequence were an inflated number in a PDF, this would be embarrassing and cheap to fix. It is neither.
Smart Bidding optimises toward whatever you have designated as a primary conversion action. Per Google’s own documentation, primary actions appear in the Conversions column and are used for bidding, while secondary actions are observation only and report into All conversions instead. Somebody had set the button tap as primary. So for sixteen months the algorithm did precisely what it was told: it went and found more people who tap phone buttons without calling.
That audience turned out to be cheap, plentiful, and largely on Display. The system was not malfunctioning. It was succeeding at the wrong objective, efficiently, with a growing budget.
Here is the part I will argue with people about: Google’s defaults make this outcome likely rather than unlikely. Click-to-call taps are offered as a conversion action out of the box, they are easy to set up, they produce satisfying volume quickly, and nothing in the interface warns you that you have just told the machine to optimise for an intention rather than an outcome. That default serves the conversion-volume story. It does not serve your pipeline.
The ten-minute check on your own account
You do not need an audit to do this. You need ten minutes and a willingness to be annoyed by what you find.
1. List what each action actually requires. Open your conversions summary and go through every action. For each one, write down the physical thing a human must do to fire it. If the answer is a tap, a page view, a scroll depth, or a video play, that is not a lead.
2. Check which ones are primary. In the conversions summary, find your goal, click Edit goal, and look at the Conversion action optimization section. Anything marked primary is steering your bidding. Anything you would not defend as a real business outcome should be secondary.
3. Check the counting setting. Google’s conversion counting documentation distinguishes Every from One. Every counts each conversion after an ad interaction, which suits sales. One counts a single conversion per click, which suits lead generation, because one person filling in your form three times is still one prospect. Lead gen accounts set to Every are inflating themselves by design.
4. Compare Conversions against All conversions. If the two columns are wildly apart, something substantial is either being ignored or being counted that should not be. Either way you want to know which.
5. Reconcile against reality. Ask the client how many enquiries they actually received last month. Not how many the dashboard claims. How many landed in an inbox or rang a phone. This single question finds more tracking problems than any tool, and almost nobody asks it.
(If step five produces an uncomfortable silence, that is your answer.)
What we did about it
Demote the button tap to secondary so it still reports but stops steering bids. Repair Enhanced Conversions so the leads that were real could be matched properly. Re-point budget toward Search, which was the only channel producing form fills, and away from the Display placements the algorithm had learned to love.
Then re-anchor bidding to the actions that survive the question in step one.
The reported conversion count fell off a cliff. That is what success looks like here, and it is worth warning a client before it happens, because a graph dropping 90% in a fortnight is alarming if nobody told you it was the plan.
What this fix cannot do
Two honest limits.
The learning history does not politely reset. Sixteen months of signal shaped which auctions the account competes in and what it thinks a good user looks like, and clearing that out takes weeks of spending into worse-looking numbers before it improves. There is no button for it.
And the leads that were never captured are gone. Fixing the measurement tells you what is happening from today. It does not recover the budget that already went to Display taps, and any agency that implies otherwise is selling you something.
Where I am still not settled
How much of those sixteen months was genuinely recoverable. It is tempting to say the account would have produced ten times the leads on a correct setup, and that would make a better story, but I cannot prove it. The counterfactual does not exist. What I can say with confidence is that the budget bought taps instead of calls, and that we can measure honestly from here.
The uncomfortable version
An account that looks fine is not evidence that it is fine. It is evidence that whatever it is measuring is going up.
Most accounts we take over have some version of this problem. Usually smaller, occasionally worse. It survives because the numbers are pleasant and nobody wants to be the person who checks.
Go and run the five steps above on your own account. If the reconciliation in step five does not match, get a free audit and we will pressure-test the tracking properly, or see what we found in other accounts. If you would rather build automation on top of your data first, read the five n8n workflows worth building, and note that every one of them assumes the numbers going in are real.
Researched and drafted with AI assistance, then reviewed, fact-checked and edited by Rajat Verma before publishing. All account examples are from real campaigns managed by Turquoise Digital and are anonymised.