Seasonality controls: stopping the weather from winning your tests

Why HVAC is harder than most industries

A retail site sees steady weekly patterns with occasional holiday spikes. An HVAC site sees demand that jumps within hours when temperatures fall, along with a shift in who arrives: in a cold snap, more emergency callers and fewer planners comparing quotes.

That mix change matters. A page designed for comparison shoppers can look worse in January simply because January visitors want a technician today. Two failure modes follow:

• False wins: a change launched just before the first frost looks brilliant.
• False losses: a change launched as the heating season fades looks like it broke the phone.

Control 1: run both versions at the same time

A true split test, where visitors are randomly assigned to version A or B on the same days, is the strongest seasonality control there is. Both versions see the same weather. This is why we prefer concurrent splits whenever traffic allows, as described under CRO experiments.

When traffic does not allow it, use a matched comparison: test areas against control areas over the same weeks, as in our holdout article.

Control 2: a demand index you did not touch

If you must compare periods, measure market demand alongside your outcome. Good HVAC demand indexes:

• Search Console impressions for broad heating queries. Google's Performance report help explains that impressions count how often your site appeared in results, that the chart aggregates by property, and that the newest data can be preliminary. Exclude the last few days before you compare.
• Calls to untouched pages, such as a maintenance page you left alone during a test on the replacement page.
• Calls from your Google Business Profile, if the change was on the website only.

A worked example (illustrative): replacement-page calls rose from 50 to 70 between two four-week periods, a 40% rise. Over the same periods, impressions for broad heating queries rose 30%. Dividing calls by the index gives 50 ÷ 1.00 = 50 versus 70 ÷ 1.30 ≈ 54. The demand-adjusted change is about 8%, not 40%. Most of the "win" was weather.

Control 3: choose your test windows

Some windows are cleaner than others. A simple planning calendar for heating-led businesses:

Climate varies enormously across Canada, so these windows shift by region. Our areas hub notes the main differences province by province.

What Google Ads seasonality adjustments are for

Google's seasonality adjustment help describes a tool for telling Smart Bidding about an expected short-term conversion-rate change, such as a promotion. Key points from that page:

• it suits short events of roughly one to seven days
• it may work less well over longer periods, especially beyond 14 days
• it is not needed when you do not expect a major change, because Smart Bidding already accounts for seasonal events
• it is available for Target CPA and Target ROAS on several campaign types, with broader support on Performance Max

For testing, the lesson is the reverse: if you apply an adjustment in the middle of an experiment, you have changed the bidding during the test. Schedule tests around planned adjustments, not across them.

Programs and prices are seasonal too

Rebate programs open, change and close, and each change moves demand for certain equipment. A heat-pump page test that straddles a program change is measuring the program as much as the page. Before launching, check current program pages for your region; the areas hub links the official sources we track, as of October 2026.

A pre-launch seasonality checklist

• Is a concurrent split possible? If yes, use it.
• If not, which demand index will you track, and from which date?
• Does the window include a forecast temperature swing or a program deadline?
• Is any bid adjustment scheduled during the test?
• Have you excluded preliminary data from the comparison?

For help designing a test around your season, call, text or email, or read how we work.