Most sites aren't big enough to A/B test.
Every CRO agency will sell you a testing programme. Fewer will tell you whether your traffic can actually support one. We'll tell you which side of that line you're on before you pay us anything.
From $5,000/month · We'll say no if your traffic can't support testing
How much traffic you need before testing works.
A/B testing is a statistics problem before it's a design problem. Splitting traffic two ways halves your sample, and a test that never reaches significance tells you nothing, it just costs three weeks. Find your row.
Nobody takes you from 0.5% to 5%.
You'll see that claim on CRO pages. It describes a tenfold improvement, which is not what optimisation does, it's what fixing something genuinely broken does, once. Here's the shape of an honest programme.
The tenfold claim
- Starting rate 0.5%
- Claimed result 5.0%
- Relative lift +900%
- Realistic? No
If a site converts at 0.5%, the problem usually isn't optimisation, it's broken tracking, the wrong traffic, or an offer nobody wants. Those are worth fixing. They're not a testing programme.
Compounding, over quarters
- Starting rate 2.5%
- Per winning test 5 to 15% relative
- Winners per quarter 2 to 4 of 8 to 12 run
- After 12 months ~3.3 to 4.0%
Less dramatic, and it's real. Most tests don't win, that's not failure, it's how you find out which assumptions were wrong before you build on them.
Illustrative figures. Your actual range depends on baseline, traffic, and how much low-hanging work is left, which is what the audit establishes.
The first audit rarely recommends a test.
On most sites there's a queue of things that are simply wrong, and no test is needed to justify fixing them. Testing starts once that queue is empty, otherwise you're measuring variants of a broken page against each other.
Four findings that come up on nearly every audit.
-
The tracking doesn't measure what you think
Duplicate conversion events, form submissions firing on page load, thank-you pages counted as goals when people bookmark them. We check this first, because every other number on the page depends on it.
-
Mobile is a different site and nobody looks at it
The desktop experience gets reviewed in meetings on desktops. Most traffic arrives on a phone, where the form is longer than the screen and the primary CTA sits below three paragraphs.
-
The form asks for more than the offer is worth
Nine fields for a PDF. A phone number required before anyone has decided they want to talk. Every field is a place to leave, and most exist because someone in sales asked once.
-
The page answers a different question than the ad
Paid traffic arrives on a page about something adjacent to what was clicked. That mismatch costs more than any button colour ever will, and it's usually free to fix.
Six things, in this order.
The sequence matters more than the list. Testing before tracking is fixed produces confident wrong answers.
CRO audit
Tracking verification first, then funnel analysis, then a prioritised list of what to fix before anything gets tested.
CRO auditTracking & measurement
Conversion events that fire once, on the right action, and reconcile with your CRM. Everything downstream depends on this being right.
AnalyticsBehaviour analysis
Heatmaps, session recordings, and form analytics. Where testing needs volume, this doesn't, which makes it the right first move on smaller sites.
Behaviour analyticsUser testing
Watching real people attempt the task. Five sessions will tell you more about a confusing checkout than a month of A/B testing at low volume.
UX testingLanding page builds
Pages built to match the campaign that sends traffic to them, rather than a homepage doing four jobs badly.
Custom designA/B testing programme
Once volume supports it: hypothesis, sample size calculated up front, and a stop rule agreed before the test runs.
Paid searchFix first. Test second.
The first month usually produces more lift than the following three, because it's spent removing things that were plainly wrong. Testing earns its place after that.
Check the tracking
Before anything else. Duplicate events, misfiring goals, and conversions counted twice make every subsequent decision wrong in the same direction.
Week 1Clear the obvious
Broken mobile layouts, forms asking for more than the offer justifies, ad-to-page mismatches. No test needed to justify any of it.
Weeks 2 to 6Structured experiments
Sample size calculated before launch, stop rule agreed in advance. Tests run to significance or get called as inconclusive, never quietly extended until they look good.
Month 2 onwardShip and iterate
Winners deploy, losers get documented, and the next hypothesis builds on what the last one ruled out. That accumulation is where the return actually comes from.
OngoingStated, so you can rule us out quickly.
From $5,000 / month
Ongoing programmes. A one-off CRO audit is scoped and priced separately if you'd rather start there.
- Flat fee, not a share of the lift. Performance-based CRO pricing sounds fair and creates an incentive to claim wins from tests that didn't reach significance.
- Every test reported, including losses. Roughly two-thirds of tests don't win. An agency showing you only winners is showing you a third of the work.
- Sample size agreed before launch. Along with a stop rule, so nobody is tempted to end a test on the day it happens to look good.
- Your tools, your accounts. Analytics and testing platforms stay in your name, with the full test history.
- We'll decline the work if testing won't help. Under 5,000 visitors a month, we'll tell you what to do instead, usually cheaper than hiring us.
Before you call.
Do I need a lot of traffic for CRO? +
How much can I realistically expect conversion rates to improve? +
How long until we see results? +
What if a test doesn't win? +
Is CRO better than SEO or PPC? +
What tools do you use? +
Will you tell us if we shouldn't hire you? +
Looking for one specific thing?
We'll check your tracking before we sell you anything.
Give us read access to your analytics. We'll verify your conversion tracking is measuring what you think it is, work out whether your volume supports testing, and tell you honestly which. Findings are yours whether you hire us or not.
Read-only access · No obligation · Findings are yours to keep