Lift = (B's conversion rate − A's conversion rate) ÷ A's conversion rate
(8.6 percent − 4.4 percent) ÷ 4.4 percent = 95 percent
What makes a test fair?
Four rules. Change one thing, so you know what caused the difference. Split people at random, so neither version gets the keener visitors. Run both versions at the same time. And decide before you start what you are counting and how long you will run, then do not stop early because one side is ahead on day three. Early leads in small tests disappear all the time.
The last rule is the one owners break. Checking every morning and stopping the moment B pulls ahead turns a test into a coin toss you keep calling until it lands your way. The entry on statistical significance covers how big a win needs to be before it means anything.
Which page books more services for a Port Macquarie mobile mechanic? (example figures)
A mobile mechanic in Port Macquarie sends Google Ads traffic to a landing page with a booking form. The owner suspects people would rather ring. Using a Google Ads experiment, half the clicks go to the current page (A) and half to a version with a large "Call now" button at the top and the form further down (B). It runs for six weeks.
| A: form first | B: call button first | |
|---|---|---|
| Clicks | 410 | 395 |
| Bookings (forms plus calls over 60 seconds) | 18 | 34 |
| Conversion rate | 4.4 percent | 8.6 percent |
| Cost per booking at $3.20 a click | $73 | $37 |
Version B nearly doubled the bookings from almost the same number of clicks, and halved the cost of each one. With 52 bookings between them and a gap of 16, the result is big enough to trust (the significance entry shows the quick check). The owner switches all traffic to B and plans the next test: whether quoting a price on the page helps or hurts.
Notice what counted. The owner measured bookings, forms and calls together, rather than form submissions alone, because B was expected to trade forms for calls. Counting only forms would have made B look like a failure.
Where can you run one without extra software?
Inside the tools you already pay for. Google Ads has experiments, which split a campaign's traffic between the original and a changed version (Google Ads Help, About custom experiments). Meta Ads Manager has an A/B test option that splits an audience between two ads or ad sets (Meta Business Help Centre, About A/B testing). Mailchimp and Klaviyo can both send two subject lines to parts of a list and the winner to the rest.
How does an A/B test fit into a CoraVision plan?
A test is a tactic, and in CoraVision each strategy becomes a project with an action plan, so a test like the mechanic's becomes steps you work through. Cora teaches those steps on your own screen, and in pill mode points at the next one while you set up the experiment yourself. Once the test is running, CoraVision reads the campaign's clicks and conversions every night, so the two sides can be compared at the weekly check-ins.
Questions
What should a small business test first?
The change closest to the money on the page with the most traffic. The offer, the headline and how people book (a form or a call button) move results far more often than a button colour. A small business only gets a few tests a year, so spend them on the big questions.
What is a multivariate test?
A test of several changes at once, in every combination, to see which mix works best. Three headlines and two pictures make six versions, and each needs enough people to judge. That needs far more traffic than most owner-led businesses have, so test one change at a time.
Can I run version A this month and version B next month?
That is a before-and-after comparison, and it cannot separate your change from everything else that moved: the season, a competitor's sale, school holidays. An A/B test runs both versions in the same weeks, split at random, so those things hit both sides equally.