Geo holdout
Withholding ads from some geographic areas and comparing outcomes against comparable areas that received them.
Rather than randomising people, randomise places. Some postcodes or metros run the campaign, comparable ones do not, and the difference in enquiries between them is the lift.
Why it matters for HVAC
Every objection to user-level holdouts falls away here. You are not trying to find two hundred thousand comparable individuals, you are trying to find enough comparable areas, and a multi-location contractor or a portfolio operator already has them.
The arithmetic is still real. You need enough separate areas to absorb local noise, four to eight weeks minimum so a single cold snap does not dominate the result, and areas genuinely comparable on demand and competition rather than merely adjacent.
How it is measured
Enquiries per thousand households in treated against untreated areas, over the test window, taken from the CRM rather than from the ad platform. Synthetic control methods improve the comparison where the areas differ, but the plain version is legible and usually enough.
A worked example
A contractor with twelve service areas runs ads in eight and withholds four, matched on population and prior enquiry rate, for six weeks. Treated areas produce 6.1 enquiries per thousand households, untreated 4.4. The lift is 1.7 per thousand, which converts into a cost per incremental enquiry that can be compared against every other channel on equal terms.
These figures are invented to show the arithmetic. They describe no real business.
The cost of running one
You give up the revenue the withheld areas would have produced for the length of the test. That is the real price, it is knowable in advance, and it buys the only honest answer available about whether the channel works.
Where this is demonstrated
The Tracking and the Conversions API pillar scores every listed agency on what they document here, and gives you the questions that tell you which one you are being sold.