Conversion rate optimization

Start by watching people fail

Conversion work goes wrong when it starts with opinions. Someone read an article about button colours, someone else thinks the hero image is wrong, and six weeks later a test has produced a result nobody can interpret.

Start with evidence instead. Session recordings of people who did not convert, filtered to the pages that matter. An hour of watching real attempts will tell you more than a month of best-practice discussion, and it is uncomfortable in a useful way. You see people miss the button entirely, hesitate at a field, scroll past the thing you thought was obvious, and abandon at a step nobody suspected.

Then the funnel arithmetic

Alongside recordings, the numbers: how many people reach each step, and what proportion continue. The biggest drop is not always where attention has been focused. Frequently the homepage is fine and the leak is on a step two clicks deeper that nobody looks at.

Segment it. Mobile and desktop often behave completely differently, and an aggregate number hides a mobile experience that is quietly failing. New and returning visitors differ. Traffic from one channel may convert at a fraction of another because of a message mismatch upstream.

Fix the obvious before testing anything

Most sites have several problems severe enough that testing them would be absurd. A form that fails validation without saying why. A phone number that is not a link on mobile. A required field asking for something people do not have to hand. Content that loads late and shifts the page while someone is reaching for a button. A checkout step that breaks on one browser.

You do not test whether to fix these. You fix them, then measure the before and after. Testing is for genuine uncertainty, not for defects.

Honesty about sample size

This is where most conversion programmes quietly waste money. Reaching a trustworthy conclusion requires a certain number of conversions, and most sites do not have them. Below a few hundred conversions a month, tests take so long that the result is contaminated by seasonality and by everything else that changed meanwhile.

If that is your situation, the honest advice is to stop testing and start improving: fix the defects, make the obvious improvements, and measure period over period accepting that attribution will be imperfect. That is less satisfying than a significance figure and considerably more useful.

What usually moves the number

In practice, the same few things recur.

Clarity above the fold: what this is, who it is for, what happens next. Sites consistently overestimate how much context a visitor arrives with.

Form friction: fewer fields, correct input types, errors shown next to the field, never clearing what someone typed.

Speed, especially on mobile. It is not a conversion tactic in the usual sense but it moves the number more reliably than most things that are.

Answering objections on the page. Price, timeline, what happens if it goes wrong, who you are. Unanswered objections do not produce questions, they produce silent departures.

Specific rather than generic reassurance. A named person, a real response time, an actual number.

What I report

Conversion rate by segment, the drop at each funnel step, what was changed and when, and the before-and-after on each change. Where a test was genuinely powered, the result and the confidence. Where it was not, I will say so rather than presenting a number that cannot bear weight.