Why most agencies cannot track your clinic's Google Ads

A pixel is a piece of code that fires when a page loads. For an online shop that is nearly enough, because the thank you page is the sale. Someone bought a jumper, the receipt page loaded, the pixel fired, done.

A patient does not book on a page view. They book inside one of fifteen different booking systems, often on someone else's domain, inside an iframe, behind a cookie banner, after a speed plugin has held back every script on the page. Or they book on a phone call, which no pixel has ever been able to see.

So the pixel was never the problem. The problem is the distance between an ad click and a real person with a real appointment, and almost nothing about that distance is standard from one clinic to the next.

I have written before about why a clinic cannot track its Google Ads bookings, and about how your booking software sets the ceiling on what you can measure. This article sits underneath both of those. Not what breaks, but what doing it properly actually takes, at scale.

The chain above is the whole argument. Between an ad click and one booking counted once there are ten links, and every one of them is a separate thing somebody had to get right.

The click has to carry a tracking id. The id has to survive the landing page. Consent has to allow the tag to run, and the script has to actually execute.

Then the booking widget has to fire a real completion rather than the widget opening. That event has to cross the iframe boundary and land on the right conversion action. That action has to count and train the bidding, and nothing else can double count it.

All ten. For one booking. Counted once.

The trap is not the number of links. It is that every one of them fails silently. No error message, no red banner, no email to anyone.

The platform keeps taking the money and keeps reporting the tag as delivered, so the account just looks quiet. Quiet is exactly what ads that are not working look like, which is why so many owners get told the wrong story about their own campaign.

They are not. The chain does not get installed once and reused. It gets rebuilt against whatever that particular clinic happens to be running, and the range of things a clinic can be running is wider than anyone expects.

We run Google Ads for over 120 Australian clinics, and across those accounts there are fifteen different booking systems. Cliniko, Nookal, HotDoc, Halaxy, Zanda, Splose, Jane, Mindbody, Calendly, HealthEngine, Pracsuite, Gensolve, Pogozo and Snapforms are the ones that come up most often. Each of them can be wired into a website in six different ways, and the way it is wired changes the entire job.

Then there is everything sitting on top of the page. The form stack. The phone measurement. The speed plugin, the consent gate, the firewall. And the politics of whose tag manager container is on the site at all.

This is why the build never transfers. You cannot lift the tracking off one Cliniko clinic and drop it onto the next, because the booking system is the only dimension that matches. The consent banner, the speed plugin, the form stack and the container situation are all still different.

One note on the phone, because two numbers below look like they contradict something I have published. Those durations are the bar for counting a call in the account. What I report to an owner as a likely booking is a higher bar again, and that is the 90 second rule.

That is 68,040 combinations.

I want to be honest about what that number is. It is straight arithmetic, not a claim that we have met all of them. Nobody has met 68,040 of anything. The point is the shape of the number, not the value.

Every new clinic that comes to us is one specific combination out of that space. You cannot find out which one by asking. The owner does not know, the web developer left three years ago, and the booking system was chosen for clinical reasons by someone who has since moved on. You find out by looking.

Here is the shape of the run we do on an account before I will put a number from it in front of an owner. Almost none of it is changing things. Most of it is proving things.

The rule underneath the whole run is that a success message is not evidence. A tag sitting in the page source is not evidence that it runs. A platform saying the change was applied is not evidence that it landed. Evidence is a read back from the account itself, or a rendered browser looking at the live page, and nothing else counts.

Even fully automated, that is roughly ninety minutes of machine work per clinic, and most of the ninety minutes is spent proving rather than changing. There is no version of this that is quick, and there is no way to skip to the end.

This is the part that is hard to say politely. When we take an account on, we grade every counted conversion in it as it arrived, before we change anything.

Roughly 58 percent are genuine bookings and calls. About 17 percent are approved proxies, which is a stand in for a booking that we count on purpose and disclose as a proxy. About 18 percent are junk counted as leads, and the remaining 7 percent we cannot classify from the outside.

So nearly one in five counted conversions, in accounts as they came to us, was not a lead. And every one of them was sitting in somebody's monthly report as a patient.

Nobody was being dishonest. The numbers were measuring the wrong events, and no part of the system was built to notice. A conversion rate over 100 percent is the loud version of this. The quiet version is an account that looks completely fine.

The payoff for fixing it is not a prettier report. It is two things. An honest denominator, so the three numbers your agency reports actually mean something. And bidding that learns from real patients, because smart bidding learns from whatever you tell it a conversion is. Ours learns from bookings. An account that counts button clicks gets better and better at finding people who click buttons.

You do not need us to see the shape of this. The calculator below takes what your account reports, asks how many of those conversions you believe are real confirmed bookings, and shows you the cost per booking underneath. That second number is the one every budget decision should be made on.

Not because they cannot understand it. None of this is clever. It is just long.

They will not do it because doing it properly costs money, produces smaller numbers to show the client, and none of the work is visible from the outside. There is no screenshot of a certified tracking chain. There is no slide for it.

I will be straight about the consequence. A competitor looking at one of our accounts sees a report with smaller numbers in it than their own report has. That is the trade, and I would make it again every time, because the alternative is charging a clinic owner to optimise toward events that were never patients.

There is one more reason we could afford to build this properly, and it is not that we are smarter. Every account we manage is a clinic, so the same fifteen booking systems keep coming up, and every combination we solve gets written down and used again.

A generalist agency meets a Cliniko iframe once a year and improvises. We have met it hundreds of times, and the answer is already on the shelf. That is what obsession with one industry buys, and it is the only way the maths of doing this ever works.

Here is the difference in practice.

If you take one thing out of this, take that sentence. The mountain is not the pixel. The mountain is proving that the number in the report is a real person who booked.

You do not have to climb it yourself. You do have to know it is there, because an agency that does not know it is there will hand you a number and call it a patient.

So ask three questions. What exact event fires when a patient finishes booking. When did somebody last open the live page in a browser and watch it fire. And what happens if it stops firing tomorrow. What you can afford to spend acquiring a patient is a maths problem, and every input to it comes off the chain in this article.