Conversion Rate Optimisation: Where to Start

Conversion rate optimisation gets sold as a testing discipline. Run experiments, watch the winners, bank the uplift. That framing works for a business with thousands of visitors a week. For the typical Malaysian service company with four hundred sessions a month, it produces tests that never reach a conclusion and a team that quietly stops believing in the numbers.

There is a second problem hiding underneath. Most sites that want a better conversion rate cannot yet say what a conversion is. The form counts, but so does the WhatsApp tap, and nobody has decided whether a phone call from the mobile header belongs in the same number. Optimising a figure nobody has defined is how three months disappear.

This guide covers what to count, where enquiries actually leak, how to choose your first change, how long a result takes to become trustworthy, and when the honest answer is that the site is not the problem at all.

PART 1 · DIAGNOSE

Why Most CRO Projects Start in the Wrong Place

IN BRIEFConversion rate optimisation starts with a definition and a leak, not a test. Decide what counts as a conversion, find the stage losing the most people, and only then change something. That sequencing is the first thing we set in most digital marketing services engagements.

Watch how the conversation usually opens. Someone says the website is not converting. Someone else suggests a new headline, a shorter form, a green button instead of a blue one. Three changes ship in the same fortnight, the enquiry count moves a little, and nobody can say which change did it, or whether the month was simply busier.

Three mistakes explain almost every stalled CRO effort:

• The conversion is undefined. Form submissions are counted, WhatsApp taps are not, and the baseline shifts every time someone adds a tracking tag.
• The change is chosen by taste. Whoever speaks loudest picks the test, so the site gets tidier without getting more persuasive.
• The traffic cannot support a verdict. Four hundred sessions a month will not separate a real improvement from an ordinary quiet week.

Only the second of those is a design problem. The first is a measurement problem and the third is arithmetic, which is why CRO so often fails before a single page is touched.

BENCHMARK BRIEFING 1 OF 4

What Counts as a Conversion for Your Business Model

IN BRIEFDifferent business models convert on different actions, and counting the wrong one makes a healthy site look broken. Pick one primary action per site, count secondary actions separately, and write the definition down where the whole team can see it.

The last row is the one that changes behaviour. Once a long-cycle business counts qualified enquiries instead of raw ones, its conversion rate drops on paper, and every optimisation decision afterwards gets better, because the number finally tracks something the sales team recognises.

PART 2 · DIAGNOSE

Where Enquiries Leak Before Anyone Touches a Button

IN BRIEFMost lost enquiries never reach the form. They leave on arrival, on the pricing question, or on the trust question. Finding which of the three costs you most is cheaper than testing, and you can do it from your GA4 reporting in an afternoon.

A conversion rate is a single number covering a journey with several exits. Treating it as one problem invites one fix, and one fix rarely lands on the exit that matters. Four exits account for most of the loss on Malaysian service sites:

• Arrival mismatch. The ad or search result promised something the page does not immediately confirm, so the visitor leaves within seconds.
• The price question goes unanswered. No figure, no range, no indication, so the visitor leaves to find a site that gives one.
• Trust runs out. No address, no faces, no proof of real work, and the reader decides not to hand over a phone number.
• Friction at the ask. Nine form fields, a required NRIC, or a WhatsApp link that opens nothing on desktop.

Speed sits underneath all four. Google’s own Core Web Vitals documentation sets the thresholds for loading, responsiveness and visual stability that shape whether a visitor stays long enough to read anything at all.

BENCHMARK BRIEFING 2 OF 4

A Modelled Leak Map of a Typical Enquiry Funnel

IN BRIEFMapping where people drop out turns one vague number into four separate, fixable problems. The biggest drop is not always the one worth fixing first, the fixable one is.

The final row belongs on the map even though it happens off the website. An enquiry answered two days later converts far worse than the same enquiry answered in ten minutes, and no amount of page optimisation recovers it. If your reply time is slow, fix that before you fix a headline.

PART 3 · DESIGN

How to Choose Your First Conversion Test

IN BRIEFChoose the first change by traffic level, not by ambition. Low-traffic sites should make obvious fixes and measure the quarter; only sites with real volume should run formal split tests. Getting this wrong wastes a quarter, which is why it belongs in the website design decision, not after it.

The honest question is not “what should we test” but “can we test at all”. Below a certain volume, a split test cannot separate a real effect from ordinary weekly variation, and running one anyway produces a confident answer built on noise.

BENCHMARK BRIEFING 3 OF 4

Modelled Effort and Impact by Type of Conversion Fix

IN BRIEFNot every fix is worth the same effort. Ranking candidate changes by likely impact against the work required stops the team spending three weeks on the item that moves least.

The bottom row is where most CRO conversations begin and where the least value sits. It is not that micro-copy never matters, it is that a business still refusing to publish a price range has larger money on the table.

PART 4 · DEPLOY

How to Run a Conversion Test You Can Trust

IN BRIEFA trustworthy test needs a written prediction, one variable, a fixed end date, and tracking you have already verified. Skip any of the four and the result becomes an opinion with a chart attached.

How to run a conversion test in six steps

Do this once properly and the format carries every test afterwards.

1. Write the prediction first. State what you expect to change and by roughly how much. A prediction you would be embarrassed to be wrong about is a prediction worth testing.
2. Verify the tracking before launch. Submit the form yourself and confirm the event fires once, not twice, the same discipline as proper Google Ads conversion tracking.
3. Change one thing. One variable per test. Two changes mean you learn nothing about either.
4. Fix the end date in advance. Google recommends letting an experiment run four to six weeks before judging it. Stopping early because the numbers look good is the commonest self-inflicted error.
5. Check quality, not just count. More enquiries at a worse fit is a loss dressed as a win. Ask the sales team before declaring victory.
6. Write down what you learned. One paragraph per test, prediction, result, decision. Six months of these is worth more than any single win.

BENCHMARK BRIEFING 4 OF 4

How Long Before a Conversion Result Is Trustworthy

IN BRIEFTime to a reliable answer depends on volume, and at low volume the answer may never arrive. Knowing that in advance saves a quarter spent watching a chart that was never going to settle.

The top row is the useful one, because it gives a small business permission to stop pretending. Under five hundred sessions a month, the sensible move is to grow traffic and fix known problems in parallel, an SEO audit usually pays better than a testing tool at that stage.

PART 5 · DRIVE

When the Website Is Not the Problem

IN BRIEFSometimes the conversion rate is low because the traffic is wrong, the offer is uncompetitive, or the follow-up is slow. Optimising the page in those cases is expensive theatre.

Four signs point away from the website and towards something the website cannot fix:

• The traffic never intended to buy. Broad keywords and cheap clicks fill a report and empty a pipeline.
• The offer is genuinely worse. If three competitors do the same job cheaper or faster, no headline closes that gap.
• Enquiries arrive and die. High enquiry volume, low closed business, the leak is after the site, not on it.
• The rate looks low because the maths is wrong. Internal traffic unfiltered, or the WhatsApp tap never counted at all.

Each of these has a different owner, and none of them belongs to the person redesigning the page. Naming the real constraint early is unglamorous and saves the most money.

FAQ

Frequently Asked Questions

1. What is conversion rate optimisation?

Conversion rate optimisation is the practice of increasing the share of visitors who take the one action your business needs. It depends on what you count, a site measuring form submissions and a site measuring WhatsApp taps are optimising different things. The work is part measurement, part design, part sales process.

2. Where should I start with conversion rate optimisation?

Start by defining your one primary conversion, then find the funnel stage losing the most visitors. It depends on your traffic level whether you can test formally or should simply fix known problems, under about a thousand sessions a month, sequential fixes beat split testing every time.

3. What is a good conversion rate in Malaysia?

There is no useful national benchmark, because the figure swings on business model, traffic source and what you count. It depends far more on intent than industry, branded search converts several times better than cold social traffic on the same page. Compare against your own previous quarter instead.

4. How long does conversion rate optimisation take to show results?

Expect six to eight weeks per change at moderate traffic, and a full quarter at low volume. It depends on how large the effect is, a page that finally answers the price question can move numbers within weeks, while micro-copy changes may never become readable. Google’s own guidance on experiments suggests four to six weeks minimum.

5. Do I need a testing tool to improve conversions?

No, most Malaysian SMEs get further without one. It depends on volume: below roughly five thousand sessions a month a testing tool adds cost and complexity without producing conclusions, so the budget is better spent fixing known problems and speeding up enquiry replies.