Conversion Rate Optimisation
Get more enquiries from the visitors you already have
Most sites do not have the traffic for valid A/B testing, and pretending otherwise wastes months. The work that actually moves conversion at normal volumes is research, friction removal and fixing paths that are plainly broken. We test when testing can answer something.
Our position
When we will and will not run a test
This is the part of CRO the industry is least honest about, so it goes first rather than in the small print.
We will run a split test when
- The page produces enough conversions for a result to resolve inside a sensible window, and we have shown you that calculation
- The change is genuinely uncertain, so the test can actually settle a disagreement
- The metric being tested is a real business outcome, not a click on the way to one
- Tracking is verified end to end before the test starts rather than debugged afterwards
We will not
- Sell a testing programme to a site whose traffic cannot support one
- Stop a test early because the numbers currently look good
- Report a result as a win when the confidence interval crosses zero
- Test something we already know is broken instead of simply fixing it
- Optimise a conversion rate upward while the quality of the enquiries falls
The maths nobody mentions
Why most CRO programmes cannot be run the way they are sold
Split testing works by detecting a difference between two groups that is larger than the noise between them. The smaller the difference you are trying to detect and the lower your baseline conversion rate, the more data you need before the result means anything. That relationship is steep, and it is not negotiable.
In practice this means a site converting at two per cent, hoping to detect a modest relative improvement, needs a volume of conversions per variant that most business websites will not reach in a quarter. Run the test anyway and you get one of two outcomes: an inconclusive result after four months, or a result you stopped early because it looked promising, which is not a result at all.
None of this means conversion cannot be improved. It means the improvement comes from a different method. At normal traffic volumes the productive work is qualitative — watching real people attempt the task, reading the funnel, finding what is broken — and then fixing things where the evidence is strong enough that testing would be a formality.
The honest sequence is: fix what is plainly wrong, clarify what is unclear, instrument what is invisible, and reserve testing for the pages and the questions where it can actually produce an answer.
Choosing a method
Which research method answers which question
These are not interchangeable. Each one tells you a different kind of thing, and using the wrong one is how teams end up confidently fixing something that was never the problem.
| Dimension | What it answers well | What it cannot tell you | Volume needed |
|---|---|---|---|
| Analytics and funnel review | Where people leave, on which device, from which source | Why they left, or what they expected instead | Low — useful from the first few hundred sessions |
| Session replay | Rage clicks, dead ends, hesitation, form fields people abandon | Intent, and whether the behaviour is typical or an outlier | Low — a few dozen relevant sessions is often enough |
| Moderated user testing | Why a path fails, in the person’s own words, while you watch | How common the problem is across your whole audience | None — five to eight participants surfaces most task failures |
| On-site surveys and exit prompts | What visitors came to do and what stopped them | Anything about the people who left without responding | Moderate — needs enough traffic for responses to accumulate |
| Customer and sales-team interviews | The real objections, the language buyers use, what nearly stopped them | Anything about visitors who never made contact | None — this is the most under-used method available |
| A/B testing | Whether one specific version outperforms another, with a confidence level | Why, and what would have worked better than either option | High — hundreds of conversions per variant, minimum |
What we find
The problems that turn up on nearly every site
- The form asks for things the enquiry does not need.
- Company size, budget range, how you heard about us, a required phone number for someone who wants an email reply. Every field is a reason to stop. Qualifying questions have a real place, but they belong after someone has agreed to talk to you, not before.
- Something is silently broken on one device.
- Validation that rejects a valid postcode format, a submit button that does nothing on an older browser, a date picker unusable on a phone. These do not show up in analytics as failures — they show up as a slightly lower conversion rate that everybody has learned to accept.
- The first screen never says who this is for.
- A visitor arriving from a search has one question: do these people handle situations like mine? A headline about excellence and partnership does not answer it. Naming the buyer, the problem and the outcome in plain words is the single most reliable copy change available.
- Nobody says what happens next.
- People hesitate before a form because they do not know what they are agreeing to. Will they be called? By whom? How fast? Is this a sales call? Stating it plainly next to the button removes a real hesitation at no cost.
- The funnel is not instrumented past the form.
- Submissions are tracked and nothing after them is. Which means nobody can tell whether the pages producing the most enquiries produce any revenue, and the whole optimisation effort is pointed at a metric that may not correlate with the business at all.
How the work runs
Instrument, observe, fix, then test what remains
Make the funnel visible
Verify what is being tracked and what is not, connect form submissions to CRM outcomes where possible, and establish which pages carry enough traffic to be worth working on at all.
You get: Verified tracking and a page-level conversion baseline
Watch real people fail
Moderated sessions with participants who match your buyer, plus session replay on the pages that matter. Most sites surface three or four significant problems in the first afternoon of this.
You get: Recorded sessions and an observed-issues log
Fix what is not in dispute
Broken paths, mobile failures, unnecessary form fields, missing clarity about who you serve and what happens next. Implemented, shipped and measured against the baseline.
You get: Shipped changes with before-and-after measurement
Test only where a test can settle it
For genuinely uncertain, high-traffic decisions, we run a properly powered test with a pre-declared metric and duration, and we report the result including when it is inconclusive.
You get: Test plan, result and an honest read of it
Review against quality, not volume
Monthly, against qualified enquiries rather than form fills. If the count rose and the quality fell, we treat that as a problem to fix rather than a number to present.
You get: Conversion review tied to CRM outcomes
Questions
What people ask before starting conversion work
How much traffic do we need before A/B testing is worth doing?
As a working rule, a page needs to produce several hundred conversions a month before a split test can resolve the size of improvement most changes actually deliver, in a timeframe short enough to be useful. Below that, tests take months and usually end inconclusive.
The exact number depends on your baseline rate and how large a change you are trying to detect. Small improvements need far more data than large ones. We work the number out for your specific pages before proposing a testing programme, and we frequently conclude that testing is not the right spend yet.
If we cannot test, how do you know a change is an improvement?
By restricting the changes to things where the evidence is not ambiguous. If four out of six people in a moderated session cannot find your pricing, adding pricing signposting is not a hypothesis that needs validating.
For anything genuinely uncertain, we say so and either leave it alone or accept the change as a judgement call and label it as one. Pretending an unmeasured change is a proven win is how CRO programmes lose credibility.
What actually gets fixed most often?
Forms, mobile behaviour and the first screen. Forms that ask for more than the enquiry warrants, error handling that fails silently, phone numbers that are not tappable, layouts that push the substance below three scrolls on a phone.
After that it is usually clarity: a page that never states plainly who the company serves, or what happens after you submit the form. Those are copy problems, and copy problems are cheap to fix.
Should we optimise for conversion rate or lead quality?
Almost always quality, measured as far down the funnel as your data allows. A page can double its form fills by removing qualifying questions and leave the sales team with the same number of real opportunities and twice the admin.
That means connecting form submissions to what happened next in your CRM. Where that connection does not exist yet, building it is usually the first piece of work, because without it every later decision is guesswork.
Do you implement the changes or just recommend them?
We implement. A recommendations document that sits in a backlog for eight months has produced nothing, and the diagnosis is the easy half of this work.
Where your development team prefers to make the changes, we write the specifications, review the implementation and verify the tracking afterwards.
Find out where your visitors are actually giving up
Send us the pages that carry your traffic and we will tell you whether your volume supports testing, and what we would look at first if it does not. If the honest answer is that the problem is upstream in your traffic, we will say that.
Related
Where to go next
- how we design and build sitesConversion logic is decided during the build, not bolted on.
- campaign landing pagesWhere testing is most often justified.
- analytics and event trackingNothing here works without a funnel you can see.
- a website that is not generating leadsThe same problem stated as an outcome.
- better-qualified enquiries
- paid media management
Last updated · Reviewed by Zubair Afzal