# What to test first when your site doesn't have much traffic > Low traffic doesn't mean you can't experiment. It means you can only detect large effects — so you have to test the things capable of producing them. - Source: https://bixlow.com/en/blog/what-to-test-first-low-traffic-site - Language: en - Published: 2026-09-29T09:00:00.000Z - Updated: 2026-10-06T09:28:18.400Z - Author: fernando - Topics: low traffic, sample size, A/B testing, new brands, minimum detectable effect, CRO, experiment design - Publisher: Bixlow --- There's a piece of advice that gets repeated to small brands: "you don't have enough traffic for A/B testing, come back when you do". It's half true, and the half that's false is the expensive one. What's true: with 300 sessions a day you will never detect a 3% relative lift. The maths doesn't allow it, no matter which tool you buy. What's false: that this means experimentation is off the table. It means a different kind of experimentation is on the table — and most small brands are running the wrong kind. ## The constraint, stated honestly The number of sessions you need per variant depends on three things: your baseline conversion rate, the size of the effect you want to detect, and how confident you want to be. Rough orders of magnitude for a 95% confidence level and a 2% baseline conversion rate: - To detect a 5% relative lift: around 120,000 sessions per variant. - To detect a 10% relative lift: around 30,000 sessions per variant. - To detect a 25% relative lift: around 5,000 sessions per variant. - To detect a 50% relative lift: around 1,400 sessions per variant. Read that list again, because it contains the entire strategy. At 300 sessions a day split across two variants, a 25% effect takes about five weeks. A 5% effect takes two years. You are not choosing between testing and not testing. You are choosing which row of that table you're going to live in. Bixlow shows you this before you launch, not after: you set the effect you're hoping for and it tells you how long the experiment will need to run. If the answer is fourteen months, you've learned something useful about that hypothesis without spending fourteen months on it. ## Test structure, not decoration The changes capable of producing 25%+ effects are almost never cosmetic. They're changes to what you're offering, to who you're offering it to, or to what you're asking the visitor to do. Things that can plausibly move a small site by 25%: - **The offer itself.** Free shipping threshold, bundle vs single unit, subscription vs one-off, a guarantee stated prominently instead of buried in the footer. - **The entry point.** Sending paid traffic to a dedicated landing page instead of the homepage. This one regularly produces effects large enough to detect in a fortnight. - **The page structure.** Whether the product page leads with social proof, with the product story, or with the specification. Not the wording — the order and the emphasis. - **Removing a step.** Guest checkout. One fewer form field. Not asking for a phone number. - **The primary action.** Whether the main CTA is "Buy now", "Add to cart" or "See sizes" changes who clicks and what they expect. Things that will not move a small site detectably: button colour, microcopy variations, moving a testimonial up forty pixels, changing "Submit" to "Send". These may well have real effects. You just can't measure them at your traffic level, and an unmeasurable effect is indistinguishable from no effect. ## Concentrate your traffic instead of spreading it The second lever is where you run the test. Most small sites spread their experiments across the whole site, which spreads the sample across pages that behave differently and makes the result mush. Run the experiment on your highest-traffic single page. For most ecommerce brands that's one or two product pages, not the catalogue. For most B2B sites it's the pricing page or one landing page that paid traffic hits. In Bixlow you scope an experiment to a URL pattern, so the test only runs where you pointed it and every session in the sample is comparable. A test that runs on one page with 200 daily sessions reaches significance faster than the same test spread over eight pages with 400 combined, because the second one is measuring eight different things at once. ## Use a wider effect window, not a lower confidence bar When the maths is tight, there's a temptation to lower the confidence threshold — run at 80% instead of 95% and call winners faster. Don't. At 80% confidence, one in five "winners" is noise, and you'll build your next six months of decisions on top of it. The legitimate lever is the minimum detectable effect. Decide in advance that you are only interested in changes big enough to matter to the business. If a 4% lift wouldn't change what you do next quarter, you don't need to be able to detect it. This isn't a compromise; it's what experienced teams do deliberately. Bixlow's results panel reports the confidence interval rather than a single number, which makes this concrete. "Somewhere between −2% and +31%" is an honest description of an inconclusive test on small samples, and it's much more useful than a green tick. ## What to do with everything you can't test You will have far more ideas than you have traffic to validate. That's normal and permanent — it's true at large companies too, just at a different scale. For the rest, use judgement and ship. Fix the things that are obviously broken without testing them: a checkout that errors on mobile Safari, a product image that doesn't load, a form that rejects valid postcodes. You don't need an experiment to tell you a bug is bad. And keep a written list of the hypotheses you couldn't afford to test. When your traffic doubles, you won't have to remember them. ## Key takeaways - Low traffic doesn't rule out experimentation, it rules out small effects. Pick the row of the sample-size table you can actually live in. - Test the offer, the entry point, the page structure and the number of steps. Not the button colour. - Scope experiments to your single highest-traffic page rather than spreading the sample across the site. - Never trade confidence for speed. Raise the minimum detectable effect instead. --- ## Meet our AI CRO agent What you just read, put to work by an agent: it understands your visitors, decides what to test or personalize, builds it on your site without a developer and tells you when a result is real. https://bixlow.com/en/ai-cro-agent