The calls your service garage’s ad account does not see
The phone rings while you are under the car. You pick up with greasy hands, hear “the clutch is slipping”, say “come tomorrow morning” and write the name in the notebook on the desk. This person will come, will pay and will take up a bay. In the ad account there is no trace of them.
The booking that leaves no trace
At a car workshop the decision is made out loud. A person with a broken-down car does not fill in a form and does not wait for a reply — they dial and ask whether they can come today. The whole booking happens in one conversation, ends up in the notebook, and from there on exists only in the day’s schedule.
So two separate documents are kept for the same thing. The notebook knows who came, what was replaced and how much was paid. The ad account knows who clicked, how much the click cost and who filled in a form. The two documents meet nowhere.
There is also a third category that enters neither of them — the calls nobody answered because everyone was under the cars. They are not in the notebook, because they did not become a booking, and they are not in the account, because they did not become an event. The ad paid for them, at the same price as the rest.
When I open a workshop’s ad account, the mismatch is visible even before I get to the campaigns. One of the two documents runs the budget. It is not the notebook.
The system learns from the examples you give it
The advertising platform does not optimise towards customers. It optimises towards the specific event you have named as the goal. Every time someone completes it, it remembers the profile behind it — device, time, behaviour on the page, what they did before the click — and sets off to find people who resemble them.
That is the whole mechanism: learning by example. The examples are the events that reach the account. A conversation that happened on the handset and ended up in a notebook is not an event for it. For the system it never happened.
That is why targeting settings weigh less than they appear to from the outside. The radius and the age range only outline where to look. Exactly whom it picks out from there is decided by the examples, not by the setting.
From then on, every further bit of spend goes towards a more precise hit on the profile it has already learned. The optimisation works, and works well. It works on a task you never consciously set it.
The person who writes and the person who calls
The two profiles differ in urgency. Someone who sends an enquiry and waits for a reply has time — the car is still running, they are interested in the price, and they are writing to other places too. Someone who dials usually has no choice: a warning light is on, there is a noise, there is a trip tomorrow.
The difference also shows in things the platform measures directly — the hour at which the page is opened, the device, how quickly the contacts are reached, which place the visit comes from. Trained only on the forms, the system gets better and better at finding people who compare.
For the workshop these are not equal people. One takes up a bay and pays the bill, the other collects quotes and often stays on their phone. The difference between them is the difference between a busy day and an empty one, and it is exactly this that is absent from the data.
That is where the feeling comes from that the ads bring only people who ask the price and vanish. It is not by chance that it brings them. They are the only type of customer for whom the account has proof that they exist.
Tapping the number is not a call
The dial button generates an event when it is tapped, and the temptation is to count it as a conversion. Between the tap and the booked customer there are several places where the person drops out: they did not dial, the line was busy, they called on a Saturday evening, they asked about a part you do not sell.
If you count taps, you train the system to find people who tap. That is closer to the truth than the form, and it is still not a booking.
The most accessible measure of the quality of a call is its duration. Arranging an appointment takes longer than the sentence “we don’t deal with that”. A threshold in seconds does not measure revenue and does not need to — it is enough to separate the calls that look like work from those that do not.
I have seen accounts in which every tap is counted as a conversion and the cost per conversion looks excellent. The number is correct. It describes calls that nobody at the workshop remembers.
How the missing signal shifts the budget
The reporting works mechanically: a result with no source is credited to nobody. The campaign that brought the calls shows spend and an empty column beside it. The campaign with the forms shows conversions at an acceptable price.
The decision that follows looks entirely reasonable — what does not deliver is stopped, and the money goes where a result is visible. The following month there are more forms, the phone rings less often and there is spare room in the schedule.
The other direction acts at the same time. The campaign that looks good gets more budget, and with the budget more examples of the same type of enquiry. The profile it searches by narrows around them and the next enquiries resemble the previous ones even more.
This cycle repeats because every single step is defensible. The mistake is not in the decision. It is that the numbers it is based on describe one half of reality and are silent about the other.
What makes the conversation visible
Three things close the chain, and they work at different levels. A separate phone number shown only to visitors from the ad makes the source known — every call on it comes from there and from nowhere else. The call itself is forwarded to the phone in the workshop, so nothing changes at reception.
The second level is the event that is returned to the account, with a duration threshold. The system gets as an example not every ring, but the ones that lasted long enough to be an enquiry. This narrows the material it learns from to something that resembles work.
The third level is the notebook. When the booked appointment, the customer who came and the bill that was paid are returned to the account as an offline conversion, the system stops training on calls and starts training on customers. That is the difference between “find me people who call” and “find me people who leave a car at the workshop”.
None of the three makes the advertising better in itself. They change the material the system learns from. From then on, the material decides.
The check you can do today
Take last week’s notebook and count the bookings that came by phone. Then open the ad account for the same days and see how many conversions it reports. The distance between the two is the part of your work that the system does not see and cannot use. I begin every workshop account from exactly this comparison, because it determines whether it makes any sense to look at the campaigns at all.
The second check is short. Open your own ad from a phone, the way a customer would open it, and tap the call button. Watch two things: whether an event appears in the account afterwards, and which number is dialled. If your main number is dialled and nothing happens in the account, the optimisation is currently running on a signal that has no connection with who walks into the workshop.
If the event is there, look at how it is defined. Does it count every tap, or is there a threshold. That is the difference between signal and noise, and it decides who you will be buying next month.
Frequently Asked Questions
Do I need special software to track calls?
The levels differ in complexity. The simplest — an event when the dial button is pressed — is set up on the website itself and needs nothing extra. A separate number with tracking and a call-duration threshold requires a dynamic number service. Feeding the customers who really came back into the account requires a regular export from the schedule or from the till system. Each next level is more accurate and more labour-intensive than the previous one.
Will the forms stop working if I start counting calls as well?
The forms stay. The change is that the account now sees two kinds of result instead of one and can weigh them. If bookings by phone turn out to be the main part of the work, they are then included in the calculation the system steers by. Until then it has no way of knowing they exist.
Won’t the customer be confused by a different phone number on the website?
The tracking number forwards to the same phone in the workshop. The person dials, it rings at your end as usual, and the conversation is the same. The only difference is in the record of who came from where. If your number is on the sign and on the company vehicles, visitors from the ad see the tracking number and everyone else sees the usual one.
Related: Advertising for car repair shops · Google advertising for a car repair shop on the day of the breakdown · Google advertising
