What is conversion rate optimization?

What conversion rate optimization means

Conversion rate optimization, usually shortened to CRO, is described by Wikipedia as the systematic process of increasing the percentage of users or website visitors who complete a desired action, such as buying a product, subscribing to a service or submitting contact details. The word that carries the weight is systematic. A redesign is not CRO, and neither is a color change someone liked. CRO is a loop: measure a step, form a guess about why people leave it, change one thing, measure again, and keep the result only if the numbers say so.

The same article traces the practice to e-commerce teams after the dot-com bust, when companies started spending more carefully on traffic and wanted the visitors they already had to do more. Testing tools that let marketers show two versions of a page to different visitors arrived around 2004, and a free tool from Google followed in 2007. Google later retired its successor, Optimize, in September 2023. The tools came and went; the loop did not change.

Wikipedia also notes two schools of thought. One leans on testing as the way to find what works. The other spends most of its effort before any test, learning who the audience is and what they need, and only then tests. A software seller with a few hundred trials a month should lean on the second school, because the traffic needed to test your way to every answer is rarely there. Research tells you which two or three ideas deserve a test at all.

The formula, and the denominator problem

The formula is one line: conversions divided by the number of people who had the chance to convert. Wikipedia gives the textbook case of 200 visitors in a day and 30 newsletter sign-ups, which is a 15% rate. The trouble is never the arithmetic. It is the choice of denominator. Divide paying customers by site visitors and you get a tiny number; divide the same customers by trials and you get a number ten times larger. Both are correct, and they answer different questions.

A trial business has at least four rates, and a useful CRO program gives each one its own name, its own denominator and its own time window.

Time windows matter as much as denominators. ChartMogul's 2026 conversion report defines free-to-paid conversion as the portion of leads or free sign-ups that turn into paying customers inside six months. That is a long window, and it is one reason published benchmarks cannot be compared to a 14-day cohort without care. Whenever you quote a rate, say what it counts, what it is out of and how long you waited.

A worked example with the arithmetic shown

Example numbers, made up for illustration: a seller starts 400 trials in a month, 60 of them pay, and the plan costs $49.00 a month. The trial conversion rate calculator gives a trial-to-paid rate of 15.0% and $2,940 of monthly recurring revenue from that cohort. Adding five percentage points would take the rate to 20.0%, which is 20 more paying customers, $980 of extra monthly revenue and $11,760 over twelve months.

Notice the phrase percentage points. Moving from 15% to 20% is five points, but it is a one-third increase in paying customers. People mix the two constantly, and a team that reports "we improved conversion by 5%" may mean either. Write points when you mean the arithmetic difference between two rates and percent when you mean growth relative to the old number.

Choosing what to fix first

The most common CRO error is working on the step that is easiest to change rather than the one that matters most. Size every candidate in customers per month, not in percent. The cheapest way to do that is to put the same funnel through two different improvements and compare the extra customers each would produce.

Take a funnel with 10,000 visitors a month, a 4% visitor-to-trial rate and a 15% trial-to-paid rate. That is 400 trials and 60 paying customers.

Both levers are plausible, and the second wins on arithmetic here because the trial-to-paid step has fewer people but each point of it is worth more. That ranking can flip with different numbers, which is exactly why you calculate it each time. Then divide by effort. A test that needs an engineer for a month and a test that needs an afternoon do not deserve equal rank even if they promise the same lift.

Research before you test

With a few hundred trials a month you cannot test your way to insight, so collect the evidence that costs nothing first. Each of the following takes less than a day and tells you which hypothesis is worth a test.

• Read every support message and cancellation reason from the last two months and sort them into a short list of themes. Count the themes.
• Watch ten recordings of trials that did not pay and ten of trials that did. Write down the first action the paying group took that the other group skipped.
• Ask each person who cancels one question, with a free-text answer, in the cancellation flow itself.
• Walk through your own sign-up on a phone, on a slow connection, with a card from a different country. Note every moment of doubt.
• List every claim on your trial page and ask whether a stranger could verify it in under a minute.

The output of research is not a to-do list. It is a ranked list of hypotheses written in a fixed form: "We believe that changing X for visitors of type Y will move rate Z because of evidence E." A hypothesis without evidence is a hunch, and a hunch is the most expensive thing to test.

Testing without fooling yourself

Wikipedia points out that statistical significance is what tells you a result is not just chance. You do not need a statistics degree to respect that idea, only a feel for how noisy small counts are. A rate measured on n people carries a standard error of roughly the square root of p times (1 minus p) divided by n. Here is that figure for a trial-to-paid rate near 15%, with the range inside which the true rate probably sits (two standard errors either side).

At 400 trials the range is about 11.4% to 18.6%, so a single month of 400 trials cannot reliably tell 15% from 18%. A five-point improvement (from 15% to 20%) is about 2.8 standard errors on that sample, which is visible but not overwhelming. This is the practical reason to test big, clear changes first and to leave small copy tweaks until your volume is high enough to see them.

• Judge a test on paid customers, not sign-ups. A variant can lift sign-ups while the extra people never pay.
• Let every cohort finish its whole trial before you read it. A 14-day trial started today reports its answer in two weeks.
• Change one thing per test, or you will not know which change did the work.
• Decide the sample size and the stopping date before you look. Stopping the moment a number looks good is how noise gets promoted to a finding.
• Record every test, including the ones that lose, with the hypothesis and the counts. A log of losers keeps you from repeating them.

Mistakes that waste a quarter

• Optimizing the rate that is easy to measure, usually visitor to trial, while the money sits in trial to paid.
• Comparing your rate to a published benchmark without matching the denominator and the time window.
• Running several overlapping tests on the same visitors and attributing the result to the wrong one.
• Treating a lift in a single week as permanent. Seasonality, a newsletter mention or a pricing change elsewhere can all move one week.
• Skipping the retention rate. A change that gets more people to pay and cancel at the first renewal has only borrowed revenue from next month.
• Calling a test finished because the dashboard shows a color, not because the intended sample arrived.

A four-week routine you can repeat

1. Week one: write down the four rates with their denominators, pull the last three months of each, and mark the weakest step by customers lost, not by percent.
2. Week two: do the research list above for that step only and turn what you learn into three written hypotheses. Rank them by evidence divided by effort.
3. Week three: build the top hypothesis, set the sample size and the end date, and launch it to a random half of new trials.
4. Week four: wait. Use the time to write down what you will do under each outcome, win, lose or flat, so the decision is made before the data arrives.

When the test ends, update the log, change the number in your funnel table and start again with the next weakest step. The trial-to-paid conversion rate glossary page defines the number you will watch most, and TrialValid's card holds for free trials describe one structural change to the trial step itself.

Sources

• Wikipedia, Conversion rate optimization, read 19 Oct 2026
• ChartMogul, The SaaS Conversion Report (Kyle Poyar, 4 Feb 2026), read 19 Oct 2026