Conversion rate optimisation (CRO) means getting more of the people who already visit your site to buy, book or enquire. It covers the whole path: product and service pages, site search, the cart, checkout and forms, not just the page an ad lands on. The work goes research first, fixes second and A/B tests third, because most sites don't have the traffic to test small changes. Guessing is expensive: in Microsoft's experiments, about two thirds of ideas didn't improve the metric they were meant to. I'm on the Central Coast, NSW, and work with businesses across Australia.
Written by Dan Picken, founder. Last checked 6 November 2026.Research first
Most CRO starts with a list of things to test. I start with where people actually drop out, then why, from your own data and from watching real people use the site. The fix often becomes obvious, and an obvious fix doesn't need a test.
Find the step that leaks
The point in the funnel where the most people drop out, by device and traffic source.
Work out why
Recordings, usability tests and what customers ask, so the fix is aimed at the real reason.
Fix, or test
Clear problems get fixed. Close calls get an A/B test, if the traffic can tell.
Worked example: two kinds of funnel
Two example businesses, each with one step losing most of its people:
An online store
Example: 1,000 carts started, 280 orders
A quote-based builder
Example: 400 quote forms started, 60 sent
Neither fix needs an A/B test to be worth doing. Test when the answer isn't obvious and the traffic can tell.
From my audits
The same patterns turn up when I look at a site's funnel. Most of them get fixed without a single test.
Add to cart, checkout steps and form starts aren't tracked, so nobody can say which step loses people. The measuring gets fixed first.
Effort goes into the home page because everyone sees it, while most people drop out on product pages or in the cart.
Delivery fees appear at the last step. Among US online shoppers who had abandoned an order, 40% said extra costs were too high, the top reason in Baymard's survey after 'just browsing'.
A version looks ahead after a few days and gets declared the winner. A month later the 'lift' has gone.
Shoppers type 'mens runners' and get no results because the products say 'sneakers'. Searches with no results are a free list of what people want.
What works for a store with millions of visits was tested on that store's customers. It may not suit yours.
How testing goes wrong
An A/B test shows two versions to similar visitors and compares what they do. Done properly, it's the most reliable way to know a change worked. These are the rules that decide whether a result means anything.
The smaller the change you want to spot, the more visitors you need, and it grows fast. Evan Miller's rule of thumb means halving the change you're looking for takes about four times the visitors.
What I do: Work out the visitors needed before the test starts, with the calculator below.
Checking a test every day and stopping the first time it looks significant can push the false positive rate to 26.1% in Evan Miller's example, when you thought it was 5%.
What I do: Fix the length up front and read the result once, at the end.
People buy differently on a Monday than on a Saturday, and around paydays. Typically, tests run in whole weeks so every day is counted evenly.
What I do: Run for full weeks, and don't start a test across a big sale unless the sale is what you're testing.
Decide what counts as a win before you start: orders, revenue per visitor or sent enquiries. Pick it afterwards and something will always look like it won.
What I do: Write the main measure down before launch. Clicks and add to carts are supporting numbers, not the verdict.
Google asks testers to show Googlebot the same thing people see, use rel=canonical on test versions, use 302 redirects rather than 301s, and remove tests as soon as they're done. It says small changes usually have little or no effect on rankings.
What I do: Set tests up that way, and take the test code out when the test ends.
Worth knowing: Google Optimize, Google's free testing tool, shut down on 30 September 2023. Testing now runs on third-party tools, so whether a test is worth running is a cost question as well as a traffic one.
A/B test traffic check
An A/B test can only spot a difference that's bigger than the everyday noise in your numbers. Put in the traffic and conversion rate for the page or step you want to test. It works out how many visitors each version needs, using the standard calculation at 95% confidence and 80% power, and how long that takes at your traffic.
A conversion here is whatever you'd judge the test on: orders, sent enquiries or bookings. Not clicks.
Time to a reliable answer
17 weeks
Too long. At this traffic, research and clear fixes will teach you more than a test.
| Visitors needed per version | 16,792 |
| Visitors needed in total | 33,584 |
| Conversions per version, roughly | 420 |
| Smallest lift a 4-week test could spot | - |
A planning guide, not a verdict on a finished test. It assumes visitors are split evenly and behave the same from week to week. Sales, seasons and ad changes add noise, so treat the weeks as a minimum, and fix the length before you start rather than stopping when the result looks good.
Consumer psychology
Friction late in a purchase costs more than the same friction early on. The research behind that, and what to do about it:
01In the cart
Kahneman and Tversky's prospect theory found people feel a loss more strongly than a gain of the same size.
A delivery fee sprung at checkout feels like losing money, not a small change in price. The loss aversion page goes further.
What I do: Show delivery costs early, on the product page, and frame a free delivery threshold as money kept.
02Choosing what to buy
In the well-known jam study, a bigger display sold less. But when Scheibehenne and colleagues pooled 50 experiments, the overall effect came out close to zero.
Cutting the range isn't a rule. Making the choice easy is.
What I do: Add a clear pick for each type of buyer, and filters that match how people actually choose. See choice overload.
I'm a marketer, not a psychologist. These are published studies, linked so you can read them, applied to funnels the way I'd apply them to yours.
Where the fixes come from
A/B testing is one of five. On a site with modest traffic, the other four do most of the work.
Funnels, device and browser splits, and site search terms.
Best for: Always first. It shows where to look, not why.
Usability tests, short surveys, recordings and what customers ask.
Best for: Any traffic level. The best option for smaller sites.
A structured check against published usability research, such as Baymard's checkout research.
Best for: Quick wins on sites too small to test. Treat the findings as likely, not proven.
Two or more versions, traffic split evenly, one main measure, a length set in advance.
Best for: Pages with enough conversions to tell. The calculator below checks yours.
Make the change and compare periods, allowing for seasons and sales.
Best for: Low traffic and clear fixes. Weaker evidence, so compare against the same weeks last year as well.
How it works
I look at where your site loses people and send the top 3 fixes. Then a 20-minute call.
Every step tracked, from first page to order or sent enquiry, by device and traffic source.
Recordings, usability tests and customer questions. Clear problems get fixed straight away.
A/B tests where the traffic can tell, with the length and the main measure set in advance.
What you get
Signals this works on
Out of the 8 signals every page and ad sends.
Landing page work is about the first page someone sees after an ad or a search. CRO covers everything after that too: product pages, site search, the cart, checkout and forms. See landing pages for the first part.
It depends on your conversion rate and the size of change you want to spot. As an example, a page converting at 2.5% needs roughly 17,000 visitors per version to reliably spot a 20% lift, at 95% confidence and 80% power. The calculator on this page works it out for your numbers.
There isn't one number that means much. It moves with industry, price, device and where the traffic came from. Compare against your own past, split by device and traffic source. For context, Baymard's average cart abandonment rate is 70.22%, across 50 studies: most carts don't become orders, anywhere.
Not if it's set up the way Google asks: no different version for Googlebot, rel=canonical on test versions, 302 redirects, and tests removed once they're done.
Use research instead: five-person usability tests, recordings and customer questions. Fix the clear problems, then compare before and after, allowing for seasons. It is weaker evidence than a test, but much better than guessing.
Since Google Optimize shut down in September 2023, testing runs on third-party tools, several of which connect to Google Analytics. Pick the tool after checking you have the traffic to test at all.
Copy, layout, forms and test set-up, usually yes. Bigger build changes go to your developer or platform with a written brief. If the site itself is the problem, see web design.
Takes a minute. Audits come back within 1 to 2 business days. For anything else, I'll get back to you as soon as possible.
What happens next
Keep going