Digital marketing

A/B testing in advertising: how to run tests that mean something

SEO-analytiikka ja hakukoneiden sijoituksen seuranta

A/B testing is a way to find out what actually works rather than what feels right to you. The difference is large — and a big surprise to many advertisers.

In brief

  • Test one variable at a time.
  • You need statistical significance (usually 100+ conversions per version).
  • Let the test run for at least 1–2 weeks.
  • 90 % of tests produce no significant difference — that is normal.

What to test in Meta advertising

  • Creative: image versus image, video versus still
  • Headline: benefit-focused versus problem-focused
  • CTA — “Buy now” versus “See more”
  • Audience: interests A versus B

What to test on a landing page

Setting up A/B tests and analysing the results
You need at least 100 conversions per version for a statistically significant result.
  • Headline: the main message
  • Image — the hero visual
  • CTA button: wording, colour, position
  • Form length: 3 versus 5 fields

Statistical significance

A test result is significant when the confidence level is above 95 %. In practice that means at least 100 conversions per version.

Under 30 conversions per version means you genuinely do not know what works. Do not draw conclusions.

The most common mistakes

A/B testing in advertising and statistical significance
A/B testing is a way to find out what actually works.
  • Testing several variables at once — you cannot tell which one mattered
  • Stopping too early — a three-day test means nothing
  • Declaring a “winner” when the difference is tiny
  • Not documenting the tests, so the same test gets run again a month later

Frequently asked questions

How many tests should run at once?

One significant test at a time per page or campaign. More than that splits the traffic.

How long should a test run?

At least a week, preferably 2–4. It depends not only on the number of conversions but on the variation between weekdays and weekends.

What is a good lift in an A/B test?

A 5 % lift is normal, 20 % is rare. If the difference is 100 % or more, check for errors.

External sources:

Want practical support?

Book a free 30-minute assessment. We go through your situation and work out the best next step.

Share article
Read next

Related articles

Get in touch

Shall we take the first step together?

Tell us briefly where you stand. We reply the same working day, usually within a couple of hours.

  • A free 30-minute call, no strings attached
  • You get a concrete view of where you stand
  • We say plainly if we are not the right fit

We reply the same working day. Your details go to us only.