Conversion / AB testing
AB Testing Services: Hypotheses, Tests and Honest Readouts
A test is only as good as its question. We write the hypothesis, check your traffic can answer it, set up clean measurement and tell you plainly when a result is not proof.
- Google Analytics
- Google Tag Manager
- Matomo
- Shopify
- Starts with
- A ranked hypothesis list
- Measured with
- Tag Manager and Analytics
- Speed
- From brief to quote in about a week
- Readout
- Plain words, stated limits
AB testing shows two versions of a page or element to different visitors at the same time, then compares how each performs against a chosen goal. Versai Studio plans the hypothesis, designs the test, sets up tracking and reads the result.
Deliverables
What you get
- 01
Test readiness check
We review your traffic, conversions and tracking to say whether a formal test can give a reliable answer.
- 02
Hypothesis backlog
A ranked list of test ideas, each with the reason, the expected effect on behaviour and the effort.
- 03
Test design
A written plan for each test: variants, audience, primary goal, stopping rule and what would change your mind.
- 04
Variant design
New versions of the page or element, designed in Figma and built with our web design team.
- 05
Measurement setup
Goals and events in Google Analytics and Google Tag Manager, tested before the first visitor sees a variant.
- 06
Readout and decision
A short report on what happened, how sure we are, and whether to ship, iterate or drop the change.
Process
How it works
- 01
Check readiness
We look at traffic, goals and tracking, and tell you honestly if testing fits yet.
- 02
Choose and design
We pick the most promising hypothesis and design the test, goal and stopping rule up front.
- 03
Run and monitor
We launch, watch for technical faults, and avoid peeking at results to declare an early winner.
- 04
Read and decide
We report the outcome in plain language and agree the next step with you.
A dashboard layout with shapes only, and no values. Illustrative, not client work.
Use cases
Where it fits
Landing page for ads
A team compares two headlines or offers on a page that receives steady campaign traffic.
Product or pricing page
A business tests how plans are presented, so visitors can compare options without confusion.
Lead form length
A services firm tests a shorter form against the current one, measuring qualified enquiries.
Online store checkout step
A store tests reassurance copy near the payment step, measuring completed orders.
In depth
In depth
Start with a hypothesis, not a button colour
A hypothesis states what you will change, why you think it will matter and what behaviour you expect. It comes from evidence: analytics showing a drop-off, user comments, or a clear gap between what an ad promises and what the page says. Testing random tweaks wastes traffic. Testing a well-reasoned change teaches you something about your visitors even when it loses. Write it as a sentence beginning with because, so the reasoning is on record.
Choose one primary goal
Every test needs one main measure, decided before launch. If you watch many metrics and pick whichever looks good, you will find false winners. Choose the outcome tied to business value, such as completed bookings, and treat other numbers as diagnostics. Make sure the goal is tracked accurately, which is why we review how results are counted first. Resist adding extra goals midway just because a different number is moving.
Traffic decides what you can learn
A test must show a difference larger than normal day-to-day noise. Low-traffic pages need big changes or long runs, and some cannot be tested formally at all. In that case, we favour careful before-and-after comparison, user feedback and sound judgement, and we say clearly that these methods prove less. Better a modest honest conclusion than a confident wrong one. Remember that a quiet week for your business is also a quiet week for your test.
Avoid the common traps
Ending a test the moment a variant leads is the classic error, because early swings are mostly noise. Others include running tests during unusual periods, changing a variant mid-test, and ignoring that visitors may see both versions. Technical flicker, where the original appears briefly before the variant, can also bias results. We check the setup before launch and watch for faults while it runs. Test the setup first by sending yourself through each variant and checking the events.
Learning beyond the winner
A test log is a long-term asset. Record the hypothesis, the setup, the result and what you decided. Over time, patterns appear: your audience responds to clarity, to proof, or to a lower commitment. Those patterns guide design choices beyond any single page. This sits within our wider conversion optimization work and informs future web design decisions. Share losing tests too, since teams learn as much from them as from winners.
AB test, multivariate test and before-and-after
| Criteria | AB test | Multivariate test | Before and after |
|---|---|---|---|
| What it compares | Two versions, shown at the same time. | Several elements changed in combination. | The same page across two time periods. |
| Traffic needed | Moderate, depending on expected effect. | Much more, since combinations multiply. | Low, but conclusions are weaker. |
| Reliability | High when run correctly. | High if traffic allows, easy to misread. | Low: seasons and campaigns also change. |
| Effort | Moderate: one variant to build and track. | High: many variants to design and check. | Low: change, then compare. |
| Good fit | One clear hypothesis on a busy page. | Very high-traffic pages with several open questions. | Small sites that cannot support a formal test. |
Scroll sideways to see every column.
Key terms
- Hypothesis
- A testable statement: if we change this, then that behaviour will shift, because of this reason. It is the starting point of every worthwhile test.
- Control and variant
- The control is the current version. The variant is the changed version. Visitors are split between them so results can be compared.
- Statistical significance
- A measure of how unlikely a result is to be random noise. It does not prove a change will keep working, or that the effect is large.
- Sample size
- The number of visitors a test needs before a result can be trusted. It depends on current performance and the smallest change worth detecting.
- Novelty effect
- A short-term bump caused by something being new rather than better. It fades, which is why tests should run through full cycles.
- Flicker
- A brief flash of the original page before a variant appears. It can distort behaviour and bias test results.
Before you launch a test
- Write the hypothesis and the reason you believe it.
- Pick one primary goal and confirm it is tracked correctly.
- Estimate whether your traffic can detect the change you hope for.
- Decide the stopping rule and the runtime before you start.
- Check both versions on a phone and in several browsers.
- Confirm visitors see one version consistently across visits.
- Avoid launching during sales, holidays or other unusual periods.
- Agree in advance who decides what happens with the result.
FAQ
Questions
What is AB testing?
AB testing compares two versions of a page or element by showing each to a different group of visitors at the same time, then measuring which one does better against a goal you set beforehand.
How much traffic do I need?
It depends on your current conversion level and how big a change you hope to detect. Small sites often lack the volume for reliable tests. We assess this first and suggest alternatives if needed.
Which testing software do you use?
We do not name a testing platform here, since none is confirmed in our stack. We agree the tool with you at scoping, and set up goals in Google Analytics and Google Tag Manager.
How long does a test run?
Long enough to cover full weekly cycles and reach the sample your plan requires. We set the stopping rule before launch and do not end a test early because one variant looks ahead.
What if the test shows no winner?
That is a result too. It means the change did not matter much, or the test could not detect it. We record what you learned and use it to choose the next hypothesis.
Can you test on a small site?
Often not in a strict sense. With little traffic, a test takes too long to be reliable. We then suggest bolder changes, careful before-and-after tracking, and user feedback, and we label those conclusions as weaker.
Related services
Read next
Read next
- BlogEcommerce Marketing Strategy: How to Build One Step by Step
- BlogAI Website Redesign: What AI Can and Cannot Do for an Online Store
- BlogCreative Fatigue: How to Spot It and What to Do About It
- BlogCreative Strategy: A Framework for Ads That Keep Performing
- BlogHook Rate: What It Measures and How to Improve It
- BlogPerformance Creative: How to Iterate Ads Faster Without Losing Quality
- BlogWhat Is a Creative Strategist? Role, Skills and How to Work With One
Get started
Start with AB Testing
Tell us about your brand and your next campaign. We reply with a direction, a scope and a timeline.