In most small and mid-sized companies in French-speaking Switzerland, website decisions get made in meetings. Someone thinks the button should be more visible, someone else prefers a different headline, and the version that wins is the one argued by the most convincing person in the room. A/B testing exists precisely to replace that discussion with a measurement. But applied without method to a site receiving 900 visitors a month, it mostly produces expensive illusions: 40 % gains that evaporate the following month, tests stopped too early, conclusions drawn from thirty clicks. This guide explains how to test usefully when you have neither the traffic nor the budget of a large online retailer.
What A/B testing actually does
The principle is simple: two versions of the same page are shown at random to two halves of your visitors, during the same period, and you compare one precise goal, for example a quote request, a phone call or a purchase. Because both groups experience the same conditions at the same time (same commercial weather, same campaigns, same season), the observed difference comes in principle from the tested change alone. That simultaneity is where the whole value of the method lies.
What A/B testing does not do: it never tells you why. It does not create traffic. It does not rescue a vague offer, a shaky price position or a form that breaks on mobile. A test compares two options you imagined; if both are mediocre, the winner stays mediocre. Before testing, you need a technically sound website, reliable conversion tracking and a clear idea of what you are selling. The foundations of the site always come before fine-tuning.
What is worth testing
Starting rule: only test what can change a buying decision, not what changes an aesthetic impression. On an SMB website, the elements that produce genuinely measurable differences are almost always the same ones.
- The main promise, meaning the headline of the homepage or landing page. Moving from Tailored solutions for your business to IT support in Geneva in under 4 hours is not a cosmetic change: it changes the perceived offer.
- The call to action and the implicit promise it carries. Get a free quote within 24 hours does not commit a visitor the way Contact us does.
- Form length. Going from nine fields to four is one of the rare changes that often wins, especially on mobile.
- Showing the price. Displaying a range, for instance from 2,500 CHF, instead of nothing at all changes the quality of incoming requests, sometimes more than their quantity.
- Proof: customer reviews, local references, guarantees, stated lead times, certifications.
- Page structure: block order, featuring one single service instead of six.
What is almost never worth the cost of a test
- Button colour, corner radius, the choice between two shades of grey.
- Micro-variations in wording, such as Send versus Submit.
- Changes located far below the fold on mobile, where most Swiss traffic actually happens.
- Anything on a page seen by fifty people a month.
The traffic question: how many visitors do you need?
This is the point that kills most SMB testing programmes. The number of visitors required depends on three parameters: your current conversion rate, the size of the gain you want to detect, and the level of certainty you demand.
The order of magnitude to remember is this: the lower your conversion rate and the smaller the gain you are chasing, the more visitors you need, and the relationship is not linear, it explodes. At a 2 % conversion rate, reliably detecting a 10 % relative gain typically requires several tens of thousands of visitors per variant. Detecting a 50 % gain requires far fewer, in the range of a few thousand. A company whose site gets 1,000 visitors a month and converts at 2 % produces 20 conversions monthly: at that scale, no honest test of a micro-change is possible.
The practical consequence is actually liberating: if your traffic is modest, only test radical changes. Two genuinely different pages, not two variants of the same headline. A large difference can be detected with little data; a 3 % difference will never be detected on your site, whatever the tool displays.
Always count conversions, never visitors. A reasonable field benchmark: below roughly one hundred conversions per variant over the life of the test, treat the result as an indication, not as proof.
How long to run a test
- Two full weeks minimum, even if the tool declares a winner after three days. Monday morning behaviour is not Saturday evening behaviour, and a Swiss B2B company concentrates most of its useful traffic on weekdays.
- Whole cycles: stop the test on the same weekday you started it, so you do not overweight a weekend or a public holiday.
- Six to eight weeks maximum. Beyond that, cookies get lost, visitors switch devices and the commercial context drifts.
- Set the duration before launching and write it down. That is the only protection against the temptation to stop the test at the exact moment it proves you right.
Tools an SMB can afford
The market has changed a lot: several long-standing solutions have disappeared or repositioned towards enterprise accounts, with annual price tags of four or five figures. For a smaller company, three families remain realistic.
- Modules built into your platform. Many CMS, site builders and online store solutions now include native testing in the subscription. That is the most rational starting point.
- Self-service specialist tools. Expect an order of magnitude of 50 to 400 CHF per month depending on visitor volume. Enough for a brochure site or a mid-sized shop.
- The do-it-yourself test. Two separate URLs, traffic split between them, and the comparison made in your analytics tool. Almost free, heavier to set up, but formidably effective when traffic comes from a Google Ads campaign you control yourself.
Two requirements are non-negotiable whatever the tool: data hosting must be compatible with your data-protection obligations, and the script must not make the page flicker on load. A test that slows the site down distorts its own result.
Reading a result honestly
This is where most smaller companies go wrong. A few simple guardrails are enough.
- Do not decide by checking results every day. Watching continuously and stopping as soon as a threshold is crossed mechanically multiplies false winners.
- Be suspicious of huge gains. A +180 % calculated on 40 conversions is almost never real. On an already decent site, genuine gains usually sit between 5 and 20 %.
- Look at the interval, not just the percentage. A result reported as +15 %, with a range running from -4 % to +36 %, simply means you do not know yet.
- Measure the right thing. A shortened form generates more requests; if their quality collapses, you lost. Track the final conversion, not the intermediate click.
- Accept flat tests. A majority of tests show no difference at all. That is not a failure, it is information: this lever does not matter to your audience, move to the next one.
When traffic is too low: the alternatives
For a large share of small Swiss companies, statistical A/B testing is simply out of reach. That is fine. Less elegant and far more profitable methods exist.
- The openly acknowledged before/after test. Rebuild a page substantially, compare two comparable six to eight week periods, and accept that the measurement is imperfect. Document the change date so you can interpret the differences later.
- Session recordings and heatmaps. Watching ten people fight with your form often teaches more than an underpowered test.
- The five-person user test. Five customers or prospects, ten minutes each, two questions: what does this company sell, and what do I do to buy it?
- The concentrated paid test. Send advertising traffic to two different landing pages: you buy the volume the decision requires and you control the pace.
- Short-cycle channels: email subject lines, ad hooks, the opening lines of a message. Volume is higher there and feedback is immediate.
A five-step protocol
- Write the hypothesis in one sentence: if we display a price range on the Services page, qualified requests will increase, because hesitant visitors are blocked by budget uncertainty.
- Choose a single success metric and a fixed duration.
- Verify that conversion tracking works, on mobile as well as desktop, before launching anything.
- Launch, touch nothing, and log every external event: campaign, press mention, school holidays, public holiday.
- At the end, decide: keep it, drop it, or rerun it with a bigger change. Then record the conclusion in one single document, so you do not retest the same thing a year from now.
A realistic testing programme for a smaller company means one serious test per quarter, not fifteen per month. Most of the benefit rarely comes from the test itself: it comes from the discipline it imposes, namely defining a goal, measuring, and writing down what you learned. If you are unsure what deserves testing first, an outside look at your pages and your numbers costs less than a quarter of underpowered tests: let us talk.