What is A/B testing and when should I start doing it?

The numbers make this concrete. To detect a modest improvement with any confidence you generally need hundreds of conversions per variant, not visits. A page producing ten enquiries a month would need years, and a test stopped early will show a winner that is noise. Businesses that run underpowered tests do not merely waste the effort, they make changes based on results that were random, which is worse than not testing.

What to do instead at low volume is make the changes that are known to work rather than testing them. A clear headline stating what you do, one action rather than four, a form asking for the minimum, visible pricing or an explanation of what determines it, and proof near the point of decision. Those are established enough that testing them locally is spending months to rediscover something already known.

The other substitute is qualitative and produces answers faster. Watch five people attempt to use your site while narrating what they are thinking. That will surface confusions no test would have identified, because a test tells you which of two versions performed better and never tells you why, and at your volume the why is the actionable part.

When you do reach the volume, test one thing at a time and decide the sample size and the duration before starting rather than watching until it looks favourable. Stopping a test at the moment it shows the result you wanted is the most common way businesses convince themselves of things that are not true, and it is easy to do without noticing.

The exception worth knowing is email, where volume arrives all at once and the mechanics are built in. Testing subject lines across a list of a few thousand can produce a usable answer in a day, which makes it the one place a small business can practise this meaningfully before the website supports it.

There is a version of this that works at any volume and gets overlooked, which is testing before publishing rather than after. Show two versions of a headline or an offer to five people who resemble your customers and ask which is clearer and why. It produces no statistical confidence and it reliably catches the version that confuses people, which is the failure mode that actually costs you conversions. A page nobody understood does not need a sample size to identify.