Conversion-rate optimisation is removing the reasons people do not act: unclear offers, hidden costs, slow pages, forms that interrogate, and trust gaps. The method is hypothesis-test-verdict rather than redesign roulette — and the statistics punish impatience more reliably than anything else in marketing.
Where conversions actually die
Clarity. The visitor has three questions — what is this, is it for me, what happens next — and if the page does not answer them above the fold, nothing else on it matters. Clever headlines lose to clear ones with boring consistency, and the more sophisticated the team, the more likely the headline is clever.
Trust. No proof, no faces, no numbers with sources, and a large ask on a page that could have been built by anyone. Trust gaps are why the case-study pattern on this site is metric-led and reference-backed: proof is conversion work, not portfolio decoration. The counter-intuitive move that works: state plainly what you will not do. Honest limits make every other claim credible, because they signal someone who is not simply saying yes.
Friction. Eleven form fields where three would do. “Budget range” earns its place because it qualifies; “how did you hear about us?” is billing the visitor for your analytics gap. In Indian and Gulf contexts, add the payment expectations — a checkout without the payment method the market prefers is friction dressed as policy — and the surprise charge at the final step, which remains the most reliable cart-killer ever invented.
Speed. Every second of load time bleeds mobile visitors, and the honest test bench is 4G on a crowded tower rather than office wifi. This connects directly to the technical work: the same fixes serve both eligibility and conversion.
The discipline that separates CRO from redecorating
One hypothesis per test. Written down before starting, in the form “shortening the form to three fields will increase submissions”. Changing five things and measuring one number tells you the bundle worked; it does not tell you which part did, which means next time you carry all five forward including whichever one was harmful.
Sample size computed before, not eyeballed after. This is the rule most often broken and most expensive to break. A test that looks like a 30 per cent lift after two days is usually noise, and calling it produces a “win” that quietly fails to appear in revenue. Compute the sample size from your baseline rate and the effect you would care about, then wait — and if the required sample is unreachable at your traffic, that is information: do not test, just fix the obvious things.
Run through complete cycles. Weekday and weekend traffic behave differently, and a test that started Monday and ended Thursday has measured a slice of your audience rather than your audience.
Record verdicts, including losses. Keep, kill, learn — in a log that outlives whoever ran the test. Organisations that do not log verdicts re-run the same test every two years, because the person who remembers the answer has left.
Why calling tests early is the central sin
Because it is invisible and self-reinforcing. An early call produces a positive result, the change ships, and nobody ever audits whether revenue moved. Do that ten times and you have a folder of documented wins and a flat conversion rate, which is a genuinely confusing position to be in — and it leads teams to conclude that CRO does not work rather than that their method did not.
The ads and CRO service applies the same discipline to paid traffic, where impatience bills by the click and the temptation to call early is proportionally stronger.
What to do when you cannot test
Most businesses reading this do not have the traffic for well-powered experiments, and pretending otherwise wastes their time. The honest programme for that situation:
Fix the things that are unambiguously wrong — the form that asks eleven questions, the missing price signal, the absent proof, the eight-second mobile load. None of those need a test to justify; they need doing.
Then measure directionally and honestly. Before-and-after on a monthly cohort is not statistically clean, and it is still information as long as you describe it that way rather than as a proven lift. Intellectual honesty about the strength of your evidence is what keeps you from building strategy on a coincidence.
And instrument properly while you are there, so that when traffic does justify testing, you have a baseline to test against rather than starting from nothing.
The one number worth watching
Enquiry quality, not just enquiry count. A form change that doubles submissions while halving qualified conversations has made the business worse and the dashboard better — which is the specific failure mode of optimising the metric nearest to hand.
Tie the measurement to the outcome you actually sell: qualified conversations, proposals sent, revenue closed. Harder to attribute, slower to read, and the only version that means anything.
A worked example: the enquiry form
The most common conversion asset on a services site, and the one most reliably over-built.
The starting state, seen repeatedly: eleven fields including company size, industry dropdown, how-did-you-hear-about-us, a phone number marked required, and a message box placed last. Submissions are low and the team concludes the traffic is poor.
The hypothesis. Three fields — who you are, how to reach you, what the problem is — will increase qualified submissions, because every additional field is a reason to leave and none of the removed ones inform the first reply.
The design. Keep budget range, because it qualifies and it saves both sides a call. Drop the analytics questions, which bill the visitor for instrumentation you should own. Make the message box first and large, because a visitor with a problem wants to describe it rather than be catalogued.
The measurement. Submissions and qualified conversations, tracked separately, over a full month. If submissions rise and qualified conversations do not, the form got easier for the wrong people and the budget field needs restoring.
That is the whole method at small scale: a written hypothesis, a specific change, a metric tied to the outcome you sell, and a verdict recorded either way.
The trust section most sites omit
Almost every services site has a proof gap and does not know it, because the team already believes the claims and cannot see them from outside.
The test is simple: read your own homepage as a sceptic with a budget. For each sentence, ask what would make it checkable. “Experienced team” — how would I verify that? “Trusted by leading brands” — which ones, and can I ask them? “Fast delivery” — compared with what, measured how?
What survives that test is your actual proof. What does not survive is decoration, and decoration converts nobody while costing the credibility of the claims around it.
The strongest available substitute when you cannot name clients: publish the reasoning. Detailed methodology, honest constraints, and an account of how you would approach the problem cannot be faked by someone who has not done the work, and a serious buyer can tell the difference. That is why this site gives its methods away — it is the most persuasive proof available to a practice that cannot always name its clients.
Related reading
/blog/local-seo — generating the traffic this converts · /blog/technical-seo — the speed half of the problem · /services/ads-cro — the paid-traffic version · /blog/google-updates — why shortcut-driven traffic converts badly anyway
Questions I actually get
How long should a test run?
Until it reaches the sample size you computed before starting, and through at least one full weekly cycle — because Tuesday traffic and Saturday traffic behave differently. Stopping when the result looks good is how teams accumulate a folder of wins that never appeared in revenue.
How many things can I test at once?
One hypothesis per test unless you have the traffic for a properly designed multivariate test, which most businesses do not. Changing five things and measuring one number tells you the bundle worked, not which part did — and next time you will keep all five, including the harmful one.
What if we do not have enough traffic to test?
Then do not test — fix the obvious things and measure the before-and-after honestly as a directional signal. Low-traffic sites gain more from removing three form fields and stating the price than from an underpowered experiment that cannot resolve the difference.
Is a redesign ever the right answer?
Sometimes, when the page's structure is fundamentally wrong rather than its details. But a redesign is one enormous untested change, so treat it as a hypothesis with a rollback plan, not as an improvement by definition.
What converts best on a services site?
Specificity and proof. Named mechanisms, real constraints, honest limits, and something checkable. Ironically, the single most persuasive thing you can publish is what you will not do — because it makes everything else believable.
Where should we start?
The form and the price. Almost every B2B site asks for more information than it needs and reveals less than the visitor wants, and both of those are same-day fixes with measurable effects.