The cheapest traffic you will ever buy is the traffic you already have.
Doubling a conversion rate halves the cost of every customer, across every channel, permanently. It is the only lever in marketing that improves the economics of work you have already paid for.
- Watching real sessions
- One change at a time
- Including the losers
Most conversion advice is decoration.
The standard list, make the button bigger, add urgency, reduce form fields, is a list of things that occasionally work and frequently do nothing. It is applied without looking at where people actually stop.
The real reasons people leave a page tend to be duller and more specific. The price is not on it. The delivery time is not on it. There is no indication that anyone has bought this before. The form asks for a phone number and the visitor is not ready to be phoned. The page loads its main image last, so for two seconds it looks broken.
None of those are discovered by reading a best practice article. They are discovered by watching what happens on the page and asking people who nearly bought.
A conversion rate is not a number you improve. It is the outcome of removing specific reasons specific people left.
That means the work is always site specific, and anyone offering a generic checklist has not looked at your site.
Find the leak, then test the fix.
We start with where people drop. Analytics gives the shape of it, session recordings give the texture, and form analytics tells you which field someone abandoned on. That combination usually points at two or three concrete problems rather than twenty vague ones.
Then we form a hypothesis with a reason attached. Not “a shorter form might help” but “forty percent of abandonments happen on the phone number field, so making it optional should recover a meaningful share of them”. A hypothesis you can be wrong about is the only kind worth testing.
Then one change at a time, with enough traffic for the result to mean something. On lower traffic sites that means sequential testing and longer windows, and we will say so rather than running an underpowered split test and reporting noise as a win.
We will report tests that lost. Roughly half of them do, and that is normal.
A CRO programme with no failed tests in it has not been tested, it has been narrated.
Deliverables.
Conversion audit
Where people drop, by page and by step, with session recordings and form analytics behind each finding rather than a list of opinions.
Prioritised hypothesis list
Each one stating what we think is wrong, why, what we would change, and what result would prove it. Ranked by expected impact against effort.
Test design
The variant, the sample size needed, the duration, and the single metric that decides it. Agreed before the test runs, so nobody reinterprets it afterwards.
Implementation
The variants built properly. No flicker, no layout shift, no test that only works on desktop Chrome.
Form and checkout work
Field-level analysis, validation that helps rather than scolds, and the specific friction that makes people abandon halfway.
Page speed as conversion work
Load and interaction performance treated as a conversion factor, because it is one, and usually a larger one than copy.
Qualitative input
Short, well-aimed questions to people who did convert and people who did not. Five real answers beat a hundred guesses.
Results, including the losses
What was tested, what happened, what we concluded, and what we are testing next. Losing tests reported as plainly as winners.
How the work runs.
Baseline
Current conversion rate by page, by device and by source. Without it there is nothing to compare a result to.
Diagnosis
Analytics, recordings and form data to find where people actually stop, rather than where it is convenient to assume they stop.
Hypotheses
Specific, falsifiable, ranked. Each one with the reasoning written down so a losing test still teaches something.
Test
One variable, adequate sample, pre-agreed decision metric. No peeking and calling it early.
Implement and repeat
Winners shipped permanently, losers documented, next hypothesis started.
Straight answers.
How much traffic do I need for this to work?
For proper split testing, enough conversions for a difference to be detectable, which in practice means a few hundred a month as a rough floor. Below that we work sequentially: change one thing, measure over a longer window, and be honest that the confidence is lower.
What we will not do is run a two week test on thirty conversions and tell you we found a twenty percent uplift.
How long does a test take?
Usually two to four weeks, driven by your traffic rather than by our schedule. Stopping a test the moment it looks positive is the most common way CRO programmes produce imaginary results.
What if the test loses?
We revert it and report it. About half will. The value of a losing test is that it eliminates a hypothesis, which narrows where the real problem is.
Can you do this on my existing site or does it need rebuilding?
Existing site, in almost every case. If the platform makes clean testing impossible we will tell you, but that is rare and it is not a reason to sell you a rebuild.
Is this the same as redesigning the page?
No, and the distinction matters. A redesign changes everything at once, so when the number moves nobody knows which change did it. Testing changes one thing so the answer is attributable.
Where this connects.
Tell us what’s next.
Send over the problem and the numbers you have. We will tell you honestly whether we are the right people for it.