Budgeting around the season
HVAC demand is not seasonal in the mild sense that retail is seasonal. It collapses and spikes, and a flat monthly budget is wrong in both directions.
The shape of the year
Two spikes and two troughs. The cooling spike arrives with the first sustained heat, the heating spike with the first hard cold, and between them sit shoulder seasons where nothing breaks and nobody is thinking about their furnace. The exact dates move by climate zone and by year, which is the part that makes planning hard.
A flat budget buys the same impressions in April as in July. In July that is a ceiling on a market that would have taken more. In April it is money spent reaching people with no problem to solve.
What the auction does to you in a spike
Every competitor in your metro increases spend in the same week, because the same weather told all of them to. Cost per thousand impressions rises for everyone, and an account that raises budget on the day the heat arrives is bidding into the most expensive auction of the year.
The counter-move is to be present before the spike. Warm audiences built in the shoulder season cost less to reach, and retargeting a homeowner who watched a video about system age in May is materially cheaper in July than finding a new one.
The learning-phase trap in a trough
Cutting a budget hard in the shoulder season feels prudent and frequently is not. Below roughly fifty optimisation events per ad set per week, ad sets re-enter the learning phase and deliver erratically, so a reduced budget buys worse results per dollar as well as fewer of them.
If the budget has to come down, it is usually better to run fewer ad sets properly than to run all of them on starvation rations.
What good seasonal management looks like
A plan written before the season rather than a reaction to the weather. Budget steps decided in advance, offers that change with the season rather than a year-round discount, and creative shot in the quiet months when there is time to shoot it.
Raising budget in steps rather than multiples. A large jump re-enters learning and can undo the thing you were scaling, at exactly the moment you can least afford instability.
A worked example
A contractor runs $4,000 a month flat, all year, so $48,000. Reshaped, the same total might run $2,000 through the shoulder months building warm audiences and shooting creative, and $7,000 through the two spikes with the audiences already built. Same annual spend, different distribution, and the spike months start with a warm audience rather than a cold one.
These figures are invented to show the shape. They describe no real business.