Brand launch: what to test in your first month online

Picture the first month of a new brand from an experimenter's point of view. There is barely any traffic, nothing to compare against, no history at all, and the customer you're selling to is still a guess. You could hardly design worse conditions for testing.

So for most of that month, our advice is to leave A/B tests alone. Use the time to get ready, so that the day you can test you're not starting from scratch, and spend the rest of it on questions that don't need statistical power to answer.

Six months from now you'll want to know whether conversion has improved since launch. If you didn't measure from the start, nobody will ever be able to tell you, and that gap is what ends up costing the most.

Get analytics working properly first. Then add your custom events, meaning the two or three behaviours you suspect predict a purchase. Install the experimentation script now as well, even though it will sit idle for a while; adding it later means a deploy at a moment when you'll have other things on your plate.

From day one, write your numbers down every week: sessions, conversion rate and average order value, split by device and by source. It's a dull spreadsheet. Six weeks later it's the baseline that lets you read everything else.

At 50 sessions a day, no experiment is going to reach a conclusion. That still leaves plenty of useful work that doesn't depend on sample size.

You can watch the session recordings, and at that volume you can watch a good share of them. You'll find things that are simply broken: a checkout that fails in one browser, an image that never loads, a form that rejects a perfectly valid postcode. Those are bugs, and you fix them without testing anything.

You can also talk to your first twenty customers. Twenty conversations will teach you more in month one than an experiment would, especially if you ask what nearly stopped them from buying. Whatever they answer is your first real hypothesis.

And you can fix what is obviously wrong. A mobile page that takes four seconds to load needs fixing. Shipping costs that only appear at the last step need moving earlier. Nobody needs an experiment to know those hurt.

Once you're at a few hundred sessions a day, two questions deserve your whole early sample. The effects tend to be big enough to detect, and the answers shape a lot of what you do afterwards.

The first is where paid traffic should land: the homepage, a dedicated landing page, or straight onto the product. Differences of 30–50% are common here, which is well within reach even at low traffic, and the answer decides how you build every campaign for the next year. As landing page optimisation goes, it's hard to find a question with more riding on it.

The second is what your offer actually is. The free shipping threshold, a bundle against a single unit, subscription against one-off purchase, a guarantee shown prominently. A new brand gets its biggest effects from offer changes, because they change the deal the customer is being offered rather than how the page looks.

Both questions are about how the business works, both show up at low volume, and both give you answers you'll reuse far beyond a single page.

Anything cosmetic: button colours, headline variants, where the testimonials sit. They may well matter, but you have no way of detecting effects that small yet, and a month's sample spent on an effect you can't see is a month's sample thrown away.

Don't test things you're going to change anyway, either. A new brand's site changes constantly during its first months, and testing an element you're about to redesign wastes the sample twice over.

The most valuable thing you can do in month one is also free: don't touch the site for two or three weeks, and record what it does.

When everything feels urgent that is hard to do. But without a stable period there is no baseline, and without one, any claim you make later about an improvement can't be checked against anything. Two quiet weeks now are what let you tell, for the rest of the year, whether your changes worked. Put the date you'll start changing things in the calendar, and until then keep filling in the spreadsheet.