A real A/B test needs around 1,500 conversions per variant. Very few sites have that — and nearly every agency sells it anyway. We do the maths for you before you pay. And if testing can't work at your traffic, we say so, and do the work that does.
How much traffic an A/B test needs isn't an opinion. It's a formula. Here it is, for a site with a 2% conversion rate.
And you never know in advance how big the effect will be. So you have to plan for the small case.
| Uplift sought | Visitors per variant | Total per test | At 5,000 visits/month |
|---|---|---|---|
| 5% | 309,928 | 619,855 | 124 months |
| 10% | 78,039 | 156,078 | 31 months |
| 20% | 19,784 | 39,568 | 8 months |
| 50% | 3,292 | 6,585 | 1.3 months |
Halve the effect you're looking for, and the traffic you need quadruples. This is why Speero — the agency behind CXL, $10,000/month — publicly turns away every client under 100,000 visitors a month and sends them to conversion research instead of testing. That is the most expensive admission in the industry, and therefore the most credible.
At 5,000 visits a month, a single test for a 10% uplift takes two and a half years. Anyone still selling you "A/B testing with statistical significance" is selling you something that cannot exist at your scale.
Run the maths on my numbers →Share of A/B tests that actually improve the target metric — at companies with in-house statisticians and millions of users.
"We have been involved in tens of thousands of A/B tests that ran at Airbnb, Booking, Amazon, and Microsoft, and have never seen any change that improves conversions anywhere near this amount."
— on a CRO case study claiming "+337% conversions". His finding on the real magnitude: "Most progress is made by small continuous improvements: 0.1%–1% after a lot of work."Not by your budget. We'll tell you which in the first call — even when the cheaper one earns us less.
Testing can't work for you — the maths above shows why. So we don't measure, we decide well: on real user behaviour, heuristics, and ten years of seeing what makes funnels fail.
You have the volume for real tests. So we run them — properly: sample size fixed in advance, no peeking, every winner replicated before it ships.
These three are standard practice in CRO. Every one of them produces numbers that look good and mean nothing.
"We'll lift your conversion by 87%" isn't a forecast, it's a sales line. Kohavi has never seen anything of that magnitude across tens of thousands of tests — realistic is 0.1–1% per winning test.
We guarantee method and transparency. Not a number nobody can know.
Checking a running test and stopping at "95% significance!" is how most agency wins are made. It pushes the true false-positive rate from 5% to 26.1%. One in four "winners" is then noise.
We fix the sample size in advance and stick to it.
From Microsoft's research: test five treatments and the false-positive rate goes from 2.5% to 12%. Six rounds of five treatments and the chance of a "significant" result exceeds 50% — even when the change provably does nothing.
That is precisely how success stories get manufactured from dead hypotheses.
In our research, not one German-speaking CRO agency published a real rate card. Some run a page titled "What does CRO cost?" that contains no figure at all. Here are ours.
All prices excl. VAT · The audit (€990) is credited in full against your first month if you continue with a retainer · Cancel monthly, no minimum term, no setup fee · Swiss clients are billed in CHF
The verifiable DACH benchmarks: €1,500–3,000/month entry (Löwenstark) and a recommendation to plan with at least €3,000/month (suxeedo). Speero in the US charges $10,000. We sit below that — for three reasons, none of which is "less quality".
No overhead. No account managers, no slide-deck department. You work with the founder.
No testing-theatre line items. We don't bill you for A/B tests that can only measure noise at your traffic. At many providers that's the single biggest item on the invoice.
Six retainers maximum. We scale through selection, not volume.
I've been building funnels meant to convert for over ten years — and I've seen enough tests that proved nothing to know when to leave them alone. The difference between an amateur and a professional in CRO isn't who tests more. It's who knows when testing is nonsense.
Numbers like that are on almost every CRO page. They're nearly always made the same way: a test with too few participants, stopped early, at the right moment. At low statistical power, a significant result systematically exaggerates the effect — it's called the winner's curse. A meta-analysis of 115 public CRO tests found: mean uplift 3.77%, only 46% significant at all, around 70% underpowered. The "winners" in it averaged nearly three times the true typical effect.
This page itself carried "87% average conversion increase" in its description until recently. That wasn't provable, and we removed it — rather than hiding it behind small print, as is customary in this industry.
What we offer instead: the maths on your own numbers, before you pay. If your traffic is enough, we test — and you get real results at honest magnitudes. If it isn't, we say so and do the work that helps without testing. What we won't do: promise you a number nobody can know.
Conversion audit for €990 — credited against your first month. You get the funnel analysis, the prioritised action list, and the power calculation on your real numbers. Afterwards you know whether testing is even possible for you. The report is yours, whether you continue or not.
No-obligation enquiry →