CRO experiments are controlled tests of a change to your website or ads, judged on qualified leads. We write the hypothesis, size the sample, set the stopping rule, build the variant and read the result with a confidence interval. If your traffic is too low for a classic A/B split, we say so and use a design that fits.
This covers the strategy, implementation and testing-and-refinement steps of the Conv.ca method. Hypotheses usually come from conversion research, and every test relies on trustworthy counts from measurement setup.
Who it is for
HVAC and heating companies with reliable lead tracking and enough eligible traffic to test bold changes. Good candidates include:
- multi-location heating companies with several busy service pages
- companies spending steadily on Google Ads, where landing page tests can run on paid traffic
- single-market companies willing to run longer, simpler tests with switchback schedules
If your site gets very little traffic, a few well-chosen switchback tests may be all that is sensible. Our article on testing with few calls explains the options.
What is included
Test design
- A one-page plan for each test: hypothesis, evidence, primary metric, guard-rail metric
- Sample size and duration worked out with the formula in our sample size article, rounded to full weeks
- A written stopping rule, including what happens if a cold snap or heat wave hits mid-test
- A seasonality check, so the test window does not straddle a rebate program change or the first frost
Build and launch
- Variants built to Google's Search Central testing guidance: temporary (302) redirects to variant URLs, canonical links back to the original, the same treatment for Googlebot as for people
- Quality checks on phones before launch, including that the call and text buttons work in every variant
- Google Ads custom experiments where the change is on the ads side, planned around Google's eligibility rules
Reading and refinement
- Results reported as an interval with a business value at each end, as in our results-reading guide
- Guard-rail check on lead quality before any rollout
- A write-up for your test archive, then the next hypothesis
What we test
Tests are chosen from evidence, but common HVAC candidates include:
- the first screen of a replacement page: offer, proof and phone number placement
- how pricing is discussed (ranges, financing mentions, "what affects the price" sections)
- the order of emergency versus planned-service messaging
- landing pages matched to ad groups versus a general service page
- the text-us option and how it is labelled
A sample test plan (illustrative)
- Hypothesis: homeowners comparing furnace replacements leave because the page never discusses price, so adding a "what affects your price" section near the top will raise quote requests.
- Evidence: call recordings where early questions are about cost; a high exit rate from the first screen on phones.
- Primary metric: quote requests and qualified calls per eligible visitor.
- Guard rail: booked visits per lead.
- Sample: worked out from four weeks of baseline, rounded to full weeks.
- Stop rule: fixed end date; a forecast cold snap extends the test by one week rather than ending it.
The plan is agreed with you before anything is built.
Timeline
Each test moves through plan, build, run and read. The run phase is set by the sample size calculation, so it ranges from a couple of weeks on a busy page to several months on a quiet one. Your proposal lists the first tests and their expected windows.
How results are measured
- Primary: qualified calls and quote requests per eligible visitor, or booked jobs where imports exist
- Guard rails: booked jobs per lead, call duration, service-area match
- Program health: share of tests that reached their planned sample without mid-test edits
We never report a lift without its interval, and we do not compare your results with outside benchmarks.
Mistakes to avoid
- Stopping at the first significant reading. It inflates false positives.
- Testing tiny changes on small traffic. The maths cannot detect them.
- Editing the page or campaign mid-test. Agree a freeze.
- Leaving a test running for months. Ship or remove it, as Google's guidance asks.
- Judging on clicks. HVAC revenue comes from booked jobs.
Plans are on pricing. To talk about your first test, call, text or email.
Frequently asked questions
How many tests will we run a year?
It depends on traffic, not on our preference. A busy multi-location site may support a test every few weeks; a small single-town company may support a handful a year. We size each one honestly and tell you before you commit.
Will testing hurt our search rankings?
Not if it follows Google's guidance: no special version for Googlebot, temporary redirects to variants, canonical links back to the original, and no test left running longer than needed. We build tests that way.
What if a test loses?
Then you have learned something without rolling out a bad change sitewide. Losing and flat tests are recorded with the same care as winners, because they stop the same idea from being tried again.