Skip to main content
Free 20-minute marketing audit
Limited spots available

Have a question?

We usually reply within 5 minutes.

WWhatsApp TTelegram

Website A/B Testing in Uzbekistan: Hypotheses, Traffic, Test Duration and Common Mistakes

How to A/B test a website in Uzbekistan: what to test first, hypotheses, sample size and duration maths, Varioqub and other tools, and low-traffic tactics.

Website A/B Testing in Uzbekistan: Hypotheses, Traffic, Test Duration and Common Mistakes

Short answer: a website A/B test randomly splits visitors into two groups: one sees the current page (A), the other a version with one element changed (B), and you measure which one produces more enquiries, calls or orders. It only works with four conditions: a clear hypothesis, a sample size calculated in advance, a duration fixed in advance, and the discipline not to stop at the first "good-looking" result. With low traffic, you rely on bigger changes and qualitative analysis instead.

Once a site is live, the next question is how to improve it. Often the design changes to suit the country manager's taste: "make the button red", "make the banner bigger". An A/B test turns that debate into numbers. It is the stage that comes after you have built a business website for Uzbekistan: with pages, analytics and the lead form in place, testing improves them step by step.

What an A/B test tells you and what it doesn't

An A/B test answers one question: "Is B better than A, or is the difference random?" Both groups arrive at the same time, from the same ads and in the same season, so the difference can be attributed to the change.

Comparing last month's old page with this month's new one is not a test: ad budgets, season and competitor prices may all have shifted in between.

What an A/B test will not do:

  • It does not explain why. Customer interviews, session recordings and conversations with your sales team do that.
  • It does not fix a weak offer. If the price or product does not fit the market, button colours will not help.
  • It does not give quick answers on low traffic. More on this below.

What to test first

You cannot test everything, so start where the impact is largest and closest to revenue. A rough order of priority:

Also read 300 Sign-Ups, 40 Attendees: Where Webinar Leads Get Lost in Uzbekistan Business · 10 min
  • 1 — What: Offer and above-the-fold headline; Why: Visitors decide within seconds whether to stay
  • 2 — What: Enquiry form: number of fields, phone format, Telegram/call button; Why: The form is the last barrier before a lead
  • 3 — What: How price is shown: "from…", packages, instalments; Why: Unclear pricing often stops an enquiry
  • 4 — What: Trust signals: reviews, guarantees, project photos; Why: A new customer does not know your company yet
  • 5 — What: Language and block order: Uzbek or Russian first screen; Why: Which part of the audience responds
  • 6 — What: Button colours, fonts, icons; Why: Usually small effects; leave them for later

Check the technical basics first. If the page loads slowly on mobile or the form fails, that is a bug to fix, not a test. See the article on website speed and mobile UX. For ideas on rebuilding the first screen, offer and form, see a landing page that sells in Uzbekistan.

In a Fergana packing room, a woman holds two dried-apricot pouch designs side by side

The hypothesis: one sentence, one variable

Write each hypothesis in this format:

If we [change X], then [metric] will change, because [observation about customers].

Example: "If we add a 'Message us on Telegram' button next to 'Get a price' on the first screen, the enquiry rate will rise, because our sales team says many customers prefer to message rather than call."

The best sources of hypotheses:

  • the questions your sales staff and operators hear most often;
  • session recordings and click maps: where people stop, what they click, where they leave;
  • form analytics: which field people abandon.

Change one thing per test. If the headline, image and form all change at once and B wins, you will not know what worked. Fix the primary metric in advance too: enquiry, call or order. Switching to "let's look at clicks" after the test spoils the conclusion.

Sample size and duration: a worked example

The number of visits you need depends on three things: your current conversion rate, the smallest difference you want to detect, and the confidence level. The standard setting is 95% confidence and 80% power. A handy approximation:

Visits per variant ≈ 16 × p × (1 − p) / d²

Here p is the average conversion rate of the two variants and d is the difference you want to detect (as a proportion).

Illustrative example. A service page gets 400 visits a day and 2% of visitors submit an enquiry. You want to detect a lift to 3%:

  • p = (0.02 + 0.03) / 2 = 0.025; d = 0.01;
  • 16 × 0.025 × 0.975 / 0.0001 ≈ 3,900 visits per variant. The exact calculation gives 3,826;
  • for two variants, ≈ 7,650 visits in total, or ≈ 19 days at 400 a day. Rounded up to full weeks: 3 weeks.

Now take a smaller difference: from 2% to 2.5%. You need ≈ 13,800 visits per variant, ≈ 27,600 in total. At 400 a day that is 69 days, more than two months. The lesson: with modest traffic you can only detect large differences, so testing small tweaks is pointless.

Duration rules:

  • at least one or two full weeks, even if the sample fills sooner;
  • Ramadan, Navruz, New Year, the run-up to the school year on 1 September and big sale days all change behaviour. If a test ran during one of these, re-check the result in a normal period;
  • keep ad budgets and traffic sources stable while the test runs.

An online sample-size calculator works too; just write the number down before the test starts.

At a Samarkand shop entrance, an assistant counts visitors with a hand tally counter on a rainy evening

Which tools to use in 2026

Google Optimize is gone: according to Google, Optimize and Optimize 360 have not been available since 30 September 2023. Older guides still mention it, so check the date.

  • Varioqub (Yandex Metrica) — What it does: Connects to the Metrica tag. Experiment types include visual editor, redirect and flags; What to watch: There is a basic version; the paid extended version is where Yandex Cloud's listing puts the visual editor, redirect experiments and extra flags and metrics, so check that your plan covers the type you need. If your site uses a Content Security Policy, Varioqub's domains must be allowed
  • Tools that integrate with GA4 — What it does: Google's own page names integrations with AB Tasty, Optimizely and VWO; What to watch: Paid services; your head office may already hold a licence, but for a small local site they are often overkill
  • An in-house test in the site code — What it does: A developer assigns visitors to groups at random and passes the variant to analytics as a label; What to watch: Cheap and flexible, but validate the split with an A/A test

If your local site already has a Yandex Metrica tag, Varioqub is the lowest-friction starting point. To confirm the connection, its documentation suggests running an A/A experiment first: both groups see the same page and you check that no difference appears.

If variants live on separate URLs, mind SEO. Google's guidance: do not show crawlers and people different pages, add rel="canonical" pointing to the original on variant pages, use a temporary 302 redirect, and remove the test code as soon as the test ends.

What to do when traffic is low

If a page gets a few thousand visits a month, and a newly launched local site may get fewer, a classic A/B test aimed at a small difference can take months or even years. Instead:

  1. Test big changes. Not a button colour but a different offer or a page with a different structure. A large difference shows up even in a small sample.
  2. Use an intermediate goal. Instead of the enquiry, measure "click on the Telegram/call button". It happens more often, so the sample fills faster. Confirm the winner on enquiries afterwards.
  3. Pick your highest-traffic page. Usually the paid-traffic landing page or the main service page.
  4. Lean on qualitative analysis. Session Replay, click maps and scroll maps in Yandex Metrica, or the free Microsoft Clarity, show why people do not enquire. Fix obvious problems (say, a button that is invisible on mobile) without testing.
  5. Test at the ad level. Creative and copy tests often answer faster. The process is covered in ad creative testing.

The same logic applies to marketplace sellers: before changing a listing's main photo, you need a hypothesis and a measurement plan. See product card optimisation.

Common mistakes

  1. Peeking and stopping early. Illustrative example: after 1,000 visits per variant, A has 28 enquiries and B has 40. B looks "43% better", but a significance test shows that even if the variants were identical, a gap this large or larger would appear by chance about 14% of the time. The usual threshold is 5%, so this is not reliable. Do not decide until the calculated sample is reached.
  2. Testing small tweaks on too little traffic. The test drags on for months and still gives no answer.
  3. Ignoring seasonality. A variant that wins in a holiday week may lose in a normal one.
  4. Several variables at once. You have a winner but no idea why.
  5. Choosing the metric afterwards. "No difference in enquiries, but more clicks" is not a conclusion.
  6. Forgetting the test. The winner is never fully rolled out and the test code stays on the site.

Pre-launch checklist

  • ☐ Analytics and goals (enquiry, call, Telegram click) are recording correctly.
  • ☐ The hypothesis is one sentence with one variable.
  • ☐ The primary metric is chosen in advance.
  • ☐ Sample size is calculated and the duration rounded to full weeks.
  • ☐ You have checked whether holidays, Ramadan or a big promotion fall inside the test window.
  • ☐ Both variants are checked on mobile and desktop, and the form works.
  • ☐ For separate URLs, canonical tags and a 302 redirect are set up.
  • ☐ A test log exists where results and conclusions are recorded.

Conclusion

An A/B test settles website debates with numbers. Start where the impact is largest, write a one-sentence hypothesis, calculate sample size and duration in advance, and do not decide until the data is in.

Your next step: take the daily visits and conversion rate of your busiest page and use the formula above to see which difference you can realistically detect, and how long it would take.

Frequently asked questions

How much traffic does a website need for an A/B test?

It depends on your current conversion rate and the smallest difference you want to detect. For example, at a 2% conversion rate, detecting a lift to 3% takes about 3,830 visits per variant, roughly 7,650 in total (95% confidence, 80% power). Detecting a smaller difference multiplies the sample several times.

How long should a test run?

Until the calculated sample is reached, and for at least one or two full weeks, because behaviour differs between weekdays and weekends. Set the duration in advance and do not stop early, even if the result looks good. If Ramadan, Navruz or a big sales period falls inside the test window, re-check the result in a normal period.

Google Optimize has shut down — what can I use instead?

Google Optimize stopped working on 30 September 2023. If your site runs Yandex Metrica, Varioqub plugs into it; it has a basic version, while some experiment types, such as the visual editor and redirects, are listed under the paid extended version. Third-party tools that integrate with Google Analytics are another option, and a simple page-swap test can be built into the site code.

Can A/B testing hurt SEO?

Not if it is set up properly. Google's guidance: do not show search crawlers and people different pages, put rel=canonical pointing to the original on variant URLs, use a temporary 302 redirect, and remove the test code as soon as the test ends.

Want help planning your first website tests?

We can review your traffic, conversion and funnel for the Uzbek market, suggest what to test first on which page, and work out the sample size and duration. If traffic is low, we will show how to grow without classic tests.

Discuss A/B testing →

Let's work out where to start

Free consultation with a DORA specialist

Message on Telegram

Let's discuss your project

Leave your number and a specialist will get in touch.

Send a request via Telegram

Your details are used only to contact you.