Start with last year's call log, week by week
Open the 2025 call log and count inbound calls for each of the 52 weeks. Write the counts down a column. That column is the demand curve for your metro, and every budget decision this year is a decision about where on the curve the money sits.
It takes an afternoon. The log is already in the call tracking account or the field software, and most owners have never seen it laid out by week.
The curve has one blind spot. It shows the calls you received, so a week when the Local Services budget ran out on Tuesday looks like a quiet week. Check the spend against every flat week before you read it as demand.
A shop with no usable log can pull the shape from Google Trends: search interest for "ac repair" in your metro, five years deep. The Trends line is rougher than your own calls and it climbs in the same weeks, which is enough to draw the year on.
Two peaks and two shoulders
The peaks are the first sustained heat of summer and the first hard cold of the heating season. Where they land, and how much of the year they cover, depends on the metro and its market type.
Phoenix has one long peak. Cooling runs May into October, every competitor bids into it, and the heating side is small enough that February goes to selling maintenance agreements.
Tampa runs about as long, with humidity behind half the summer calls and storm weeks that empty the calendar and refill it two days later.
Columbus has two separate events with separate keywords: July heat, then the first hard freeze in November. April and late September are quiet enough that the phone rings for tune-ups and warranty work.
The shoulders are the weeks in between, and that is where a flat monthly budget does its damage, paying July prices in a week when the searches are not there.
What each channel does in each period
Local Services Ads and search campaigns follow demand week to week, and they take most of the money in the two peaks. The homeowner with no cooling is searching this minute, and the top of that results page is what you are buying.
The customer list carries the shoulders. An email and a text to people who already paid you cost almost nothing to send and fill tune-up slots in the weeks when strangers are not searching.
Meta sells the planned purchase: the tune-up, the membership, the replacement a homeowner has been putting off since the last repair. Those campaigns earn in March and September.
Website and SEO work goes into the quiet months. A service page rewritten in the middle of July loses traffic while the new version settles, and July is the month that page is worth the most.
Capacity caps the budget
When the install backlog is two weeks out and growing, replacement spend comes down and repair spend stays. A homeowner quoted for a system you cannot install until the following month calls the next company on the list, and you paid for that call.
Capacity is the question owners skip most often, and it sets what gets bought the following week. The budget number for a week is the answer to how many more jobs the trucks can take, asked separately for repairs and for installs.
How much in total
Work backward from cost per booked job and truck capacity, one job type at a time. Both numbers sit on the weekly report.
Say a booked repair costs $75 through Local Services and the trucks can take 40 more repairs in a July week. That is $3,000 of Local Services budget for that week. Say a booked replacement costs $600 and the install crews have room for 6 more that week; that is another $3,600. July is the sum of those two lines, and it holds for as long as both inputs hold.
A percentage of revenue is the usual rule of thumb. It comes from last year's books, and it does not move when the install calendar fills up.
The plan is 52 rows and the rows move on Monday
The whole thing is a spreadsheet: one row per week, one column per channel, planned spend in the cells. Print it in January and it is wrong by March.
On Monday you read three things: the 10-day forecast, the install backlog, and last week's cost per booked job by channel. Then you change the next three rows and leave the rest of the year as drawn until something moves it.