A/B testing
Some links on this page are affiliate links: we may earn a commission at no extra cost to you. Our approach
A/B testing replaces opinion with measurement. Rather than debating whether the button should be blue or orange, you show visitors two versions and look at which produces more conversions on a sufficient volume. ## A hypothesis before a variant Useful tests start with a behavioural hypothesis. "Reducing the form from seven to three fields will increase the submission rate" is a hypothesis. Changing the button colour with no idea behind it is not. ## Rigour protects against chance Without sufficient volume and without a minimum duration, a test measures noise. Natural fluctuations sometimes produce 10 % differences that disappear once the test ends. Sample calculation and respecting the duration are the boring but essential part. ## Headline and offer first The biggest gains rarely come from visual details. The headline, the promise, the proof and the form often dictate the essentials. A well-written headline test can produce more than ten button tests.When this guide is for you
- You have traffic but do not know how to improve the conversion rate
- You want to decide between two headlines, two forms or two visuals without subjective judgement
- You receive several conflicting opinions on what works
- You invest in advertising and want to test promises before scaling them
- You see a regression after a site change and want to verify it
Step by step
- Formulate a hypothesis — Write a clear hypothesis: "a shorter form will increase the quote request rate". Specify what changes, why it should work and which indicator will measure success.
- Prepare the versions — Version B changes only one element at a time: headline, form, button, image. Changing several elements at once makes it impossible to know which produced the effect.
- Calculate the sample — Estimate the traffic needed based on the current conversion rate and the target difference. A test launched on too small a sample gives misleading results; avoid stopping a test too early.
- Run and analyse — Let the test run for at least two weeks, allowing for sales cycles. Analyse statistical significance, then decide whether to deploy the winning version or test a new hypothesis.
What it costs (in MAD)
| Service | Indicative price | Good to know |
|---|---|---|
| Designing and setting up a test | 2,000 – 8,000 MAD (≈ 180 – 730 €) | Hypothesis, variant, implementation and sample calculation |
| Monthly testing programme | 8,000 – 25,000 MAD / month (≈ 730 – 2,300 €) | 2 to 4 tests per month with analysis and recommendations |
| Testing readiness audit | 5,000 – 15,000 MAD (≈ 450 – 1,400 €) | Tracking, traffic volume and friction points |
| Testing tool | 300 – 3,000 MAD / month (≈ 27 – 270 €) | Depending on traffic and features |
| In-depth statistical analysis | 3,000 – 10,000 MAD (≈ 270 – 900 €) | For a high-stakes test or a multivariate test |
Indicative ranges for the Moroccan market in 2026, excluding tax. Prices vary with scope and provider.
Mistakes to avoid
- Launching a test without enough volume, then concluding on a difference of three clicks
- Changing several elements at once, without knowing what explains a result
- Stopping the test as soon as it shows a favourable difference, ignoring normal fluctuations
- Testing visual details when the problem lies in the promise or the offer
Frequently asked questions
How much traffic is needed to launch an A/B test?
It usually takes several hundred conversions or tens of thousands of visits to get a reliable result. With less traffic, it is better to test on a campaign or emails, where the audience is more targeted.
How long should a test run?
At least two weeks, to cover the weekly cycle. Avoid tests of a few days, or those that cross a high-seasonality period.
Is a non-significant result useless?
No. It often shows that the tested element has no real impact, so effort can be focused elsewhere. The biggest gains rarely come from the button, and often from the headline and the offer.
Should mobile and desktop be tested separately?
It is better to test separately, as behaviours differ a lot. An overall result can hide an improvement on desktop and a regression on mobile.
What gain can be expected?
A regular, well-run programme often adds 5 to 20 % more conversions over six months. The strongest gains come from changing the promise, not the colour of a button.
Updated on
Related guides
Version française: /services/cro-analytics/ab-testing/ · AR · NL