You will be able to choose a sensible testing approach for a small business with modest traffic.
After lesson 7.2, Mei Ling put her own numbers into a sample size calculator. Her cake order page in Punggol gets about 150 visitors a week, and about 5 percent of them send an order request (example figures). She wanted to test a new photo she thought might lift orders a little, say from 5 to 7 percent. The calculator said she needed about 2,213 visitors per version. At her traffic, that is roughly thirty weeks. She would be testing one photo for more than half a year.
Most small businesses hit this wall. It does not mean testing is only for big companies. It means a small business has to test differently: bigger changes, in places with more traffic, and with other kinds of evidence alongside.
The sample size you need depends heavily on the size of the difference you are trying to detect. A tweak that might move a rate by a fraction needs huge numbers. A change that might double it needs far fewer.
For Mei Ling, testing a change that could take order requests from 5 to 10 percent needs about 435 visitors per version, or 870 in total. At 150 a week, that is just under six weeks, which she rounds up to six full weeks. Still slow, but finishable.
So what counts as bold? Changes a customer would notice and might change their mind over: a different offer, such as free delivery within Punggol or a small discount for orders placed two weeks ahead; a different page structure, such as putting prices and the order form at the top instead of below a long gallery; a much shorter form; a different main photo showing the cake at a real party instead of on a white background. Button colours, small wording changes and font sizes are unlikely to make a difference she could ever detect with her traffic.
The trade-off is that a bold change bundles more into one idea, so a win tells you a little less about exactly why. With small traffic, that is a price worth paying. A result you can trust about a big idea beats no result about a small one.
People behave differently on different days. Mei Ling's order requests cluster on weekday evenings and Sunday afternoons, when parents plan the week ahead. Farah's workwear shop sells more on weekdays at lunchtime. Weekend visitors may browse more and buy less, or the reverse.
If a test runs from Wednesday to the following Tuesday, it covers every day once, which is fine. If it runs from Monday to Thursday, it misses the weekend entirely, and the result describes weekday visitors only. If it runs for ten days, some days of the week count twice and others once.
So always run tests for whole weeks: one, two, six, whatever the calculation says, rounded up to full weeks. Avoid starting or ending around public holidays, school holidays or big sale periods, when behaviour shifts for reasons that have nothing to do with your change. If one of those falls in the middle of a test, note it in the write-up.
Your website may be the slowest place to test. Two other places often reach a usable sample faster.
The first is your email list. Most email tools have a built-in A/B test that sends two versions to two random parts of the list. Every send reaches all the people at once, so a test finishes in days instead of months. Mei Ling has about 2,400 subscribers (an example figure). A test of a new offer that could double the click rate from 4 to 8 percent needs about 553 recipients per version, so one send to the whole list is enough. A subject line test is easier still, though Email marketing and automation explains in lesson 8.1, Why open rates no longer tell you much, why clicks or orders make a better main measure than opens.
The second is ad platforms. Meta and Google Ads both have built-in tools for split tests and experiments that divide the audience between two versions of an ad, audience or campaign setting and report the result. Module 9 of Paid ads: Meta, Google and TikTok covers testing ads in more depth. Because ads can reach thousands of people a week, a bold ad test can finish quickly, and what you learn about which message works can then shape your website and emails.
Some questions do not need a test at all. If people are dropping out of your order form, watching a handful of real people try to use it will often show you why within an afternoon.
Ask five people who fit your customer profile, such as parents planning a child's birthday, to place a pretend order on your site while you watch, on their own phones. Ask them to think aloud. Do not help or explain. Note where they hesitate, what they misread, where they give up.
Mei Ling tried this with five parents from her block's Facebook group. Three of them could not work out whether the price shown was for the whole cake or per slice. No A/B test would have told her that, and her traffic would have needed months to show the effect. She fixed the wording the same evening.
Watching people does not prove a change works. It finds problems worth fixing, and fixing an obvious problem does not need a test. Save the tests for real choices between two good options.
With your own traffic numbers in hand, the question becomes which test you could actually finish in the next four weeks. For most small businesses that points to a bold change, run in email or ads, or no test at all and five people watched instead.
Choose one test you could realistically finish in four weeks with your traffic, and write why it is the right size for your business.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).