Ask ten marketing teams what they are doing about conversion and eight will describe a button. They moved it, they changed its colour, they tested two words of copy on it. Somewhere a dashboard reported a 4 percent lift with a confidence interval wide enough to drive a bus through, and everyone went home happy.
That is not optimisation. That is decoration. The conversion rate optimization best practices that actually move revenue tend to be duller, slower and considerably less fun to present in a deck, which is probably why so few teams bother with them.
Start with the traffic you already paid for
Most sites have a leak long before they have a persuasion problem. Before touching a headline, find out where people are giving up. Pull the funnel by step, by device and by traffic source, and look for the drop that does not match its neighbours.
The pattern repeats everywhere. Desktop converts at 3 percent, mobile at 0.6 percent, and nobody noticed because the blended number looked acceptable. Or paid social visitors bounce at twice the rate of organic ones, which usually means the ad promised something the landing page does not deliver. Fixing that mismatch is worth more than a year of button tests, and it costs nothing but attention.
This is the same logic behind good ecommerce conversion rate optimization work: plug the holes in the bucket before you turn the tap up higher.
Speed is a conversion feature, not an engineering chore
Page performance keeps getting filed under technical debt when it belongs in the revenue column. Every additional second before a page becomes usable removes a slice of your audience, and the effect is worst on exactly the traffic you overpay for, which is mobile users on ordinary connections.
You do not need a rebuild. You need to find the three heaviest things on the critical path, which are almost always an oversized hero image, a font loading strategy nobody chose deliberately and a stack of third-party tags that accumulated over four years. Cutting those usually buys more conversions than any copy change, and the improvement applies to every page at once.
Watch people use the thing
Analytics tells you where users stopped. It never tells you why. For that you need to watch someone try to complete the task while narrating their confusion, which is uncomfortable for about ten minutes and then extremely useful.
Five sessions is enough to find the big problems. Usability testing research has made this point for thirty years and it still surprises people: a handful of observed sessions surfaces most of the issues that matter, because real failures cluster rather than scatter.
What you will find is rarely subtle. A required field that nobody understands. A shipping cost that appears at the last possible moment. A trust signal buried below the fold. Those are not A/B test candidates. They are things to fix on Monday.
Test fewer things, and test them honestly
Formal conversion rate optimization testing has a statistics problem that most tools quietly encourage. Teams peek at results daily, stop the moment a variant looks good, and declare a win that evaporates the following month.
Three rules keep you honest. Decide the sample size and the run length before the test starts, and do not stop early because the line looks promising. Run for whole weeks, since Tuesday buyers and Sunday buyers behave differently. And ignore anything below roughly a 5 percent relative change unless your traffic is enormous, because you cannot reliably detect it and chasing it burns the calendar.
Most sites do not have the volume for a meaningful test programme at all. If you get 2,000 sessions a month, testing is theatre. Spend that energy on research and obvious fixes instead. The conversion rate optimization techniques that work at that scale are qualitative, not statistical.
Say the thing your customer would say
The highest-leverage copy change is almost never a cleverer headline. It is replacing your internal vocabulary with the customer's. Support tickets, sales call recordings and review sites are full of the actual words people use to describe the problem you solve, and those words outperform anything written in a brand workshop.
Language matters in a more literal sense too once you sell across borders. Research consistently finds that most global consumers will not buy in a language they do not read comfortably, and a half-translated checkout is one of the most expensive leaks a growing store can have. A machine-translated product page might scrape past for browsing. It will not survive the moment somebody has to enter card details.
Instrument before you optimise
A surprising number of optimisation programmes run for months on numbers that were never trustworthy. Duplicate purchase events, bot traffic counted as sessions, a consent banner blocking measurement for a third of European visitors. If the baseline is wrong, every conclusion drawn from it is wrong too.
Spend a week auditing the tracking before you spend a quarter acting on it. Reconcile the analytics number against the payment processor. Check that the mobile funnel fires the same events as desktop. Boring work, and it protects everything downstream.
What to do first
If you inherited a site and a mandate to improve conversion, the order is roughly this. Verify the data. Segment the funnel and find the worst mismatch. Watch five people attempt the main task. Fix what they trip over. Cut the page weight. Then, only then, start testing the things you are genuinely unsure about.
None of it makes for an exciting slide. It does tend to produce the kind of compounding improvement that shows up in the quarterly numbers, which is the only test that really counts.