Conversion work that holds up after the test finishes

The guest checkout work at JoyBuy is a fair example of how this usually goes. New-customer conversion moved from 61.94% to 68.58%, a rise of 6.64 percentage points, which worked out at roughly 383 extra orders a day and an estimated 2.24m euros of incremental value over 90 days.

None of that came from a clever idea. It came from removing a requirement that made sense internally and made no sense to somebody buying for the first time.

That's the pattern I see most often. The biggest conversion gains tend to be sitting inside a decision the business made years ago for a reason nobody can quite remember.

A single test tells you very little. A programme of tests, run often enough that being wrong is normal, tells you quite a lot.

Scaling testing at Selfridges from around three studies a year to a weekly rhythm was mostly an operational problem rather than a research one. Recruitment, scheduling, a budget that had to stretch, and getting teams to write down what they expected before they saw the result.

The £60k annual budget was a real constraint and probably a helpful one. It stopped us testing things we could have reasoned our way through and kept the effort on the questions that genuinely split the room.

Conversion rate is a useful number and a slightly misleading one on its own. You can raise it by attracting fewer people, or by pushing customers into a purchase they later return.

The Shop-in-Shop work at JoyBuy was interesting for that reason. It contributed to 14x GMV, a 35% rise in ARPU and a 5.5 percentage point rise in repurchase rate, and the repurchase figure is the one I care about most.

Somebody coming back is the clearest evidence I know that the experience was actually good rather than just persuasive.

I've learned to be careful about carrying findings between contexts. Something that lifted conversion in travel booking, where people research for weeks, often does nothing in a luxury retail environment where the decision is faster and more emotional.

Best practice lists are a reasonable place to start and a poor place to finish. I usually treat them as hypotheses rather than answers.

I also try to keep track of the tests that failed, because those are the ones that stop a team confidently repeating a mistake in eighteen months when everyone involved has moved on.