Tracking and the Conversions API
Your ad account is only as good as the events you feed it. Twenty of the hundred points here go to what each agency documents about the signal it sends back to Meta, and the gap between top and bottom comes down to one thing most buyers never think to ask about.
The distinction this whole criterion turns on
Nearly every agency in this market will tell you leads go straight into your CRM. That is lead delivery. It moves a name and a phone number from a form into the software your office manager works in, which is useful, and which changes nothing about how Meta buys your next impression.
Sending a booked job back to Meta is a different operation. When Tuesday's form fill becomes Friday's scheduled install, that outcome is uploaded as a conversion event, matched to the original person by hashed email or phone number. Meta then knows which of the people who filled in your form turned into work, and starts hunting for more of those rather than more form-fillers. The same word, CRM, appears in both descriptions. The effect on what you pay per booked job is not remotely the same.
One agency in this set documents a Conversions API implementation at all. None documents sending booked-job events back. That is why the highest score on this criterion is 9 out of 20.
Why the browser pixel alone stopped being enough
The pixel runs in the homeowner's browser, so it can be blocked by an ad blocker, held back behind a cookie banner, or trimmed by the browser itself. Events that never fire are events Meta never learns from. The Conversions API sends the same events from your server, where none of that applies, and pairs them with the browser events through a shared event ID so one action is not counted twice.
The number that tells you whether it is working is event match quality, which Meta scores from 0 to 10 on how much identifying information each event carries. HVAC has an unusual advantage here, because your leads arrive with a phone number attached and a phone number is one of the strongest match keys there is. An agency that has never mentioned this score to you is probably not looking at it.
Five questions for the call
- Are you sending events from the server, or only from the browser pixel?
- When a lead becomes a booked job, does that go back to Meta, and how?
- What is our event match quality score, and what is it on phone numbers?
- How are pixel and server events deduplicated?
- Which event are we optimising toward: form fill, booking, or something further down?
The last one matters more than it sounds. An account optimised toward form fills will reliably find you people who fill in forms, and some markets are full of people who will fill in a form and never pick up the phone.
How the nine score here
Out of 20, from each agency's own published pages, read 12 September 2026. Full band definitions on the method page.
| Band | Score | What is documented | Who |
|---|---|---|---|
| Basic | 4 / 20 | Pixel and standard events, with leads delivered into a CRM. Nothing documented flowing back to Meta. | |
| Not scored | n/a | Listed but not scored. See the note on the ledger. | Ascenditt |
Seven of the eight scored agencies document exactly the same thing, which is why they share a row. One is ahead of the field on the criterion that carries twenty of the hundred points, and it is ahead by documenting something the others have not written down.
It scores what each agency publishes, because that is the only thing about a third party anyone can check. An agency running immaculate server-side tracking that never wrote it down on a web page scores badly here. If you have one of them on a call, ask the five questions above and believe the answers over this column.