Short answer
To increase conversion rate, first confirm that conversion tracking works, then find the funnel step with the largest loss, research why people leave there with behaviour data, and put the change into an A/B test with a written hypothesis and a sample size worked out in advance. Out of order, the test is spent on bad data.
In the Monday meeting someone points out that the conversion rate is low, and someone else says, “let’s make the button on the home page orange.” Two days later the button is orange, three weeks later the report says “up a little”, and nobody knows whether the rise came from the button, the campaign or the season. How to increase conversion rate is not answered by a list of tips; it is a job with an order: measure first, then find where people are lost, then understand why, and test last. This article follows that order step by step.
A quick definition: conversion rate is the number of conversions divided by the number of sessions. A conversion is the action that is worth something to you: an order, an enquiry form, an appointment or a phone call. To see your own figure, use the conversion rate calculator; the steps below explain how to raise it.
1. Check the measurement first
If conversion tracking is wrong, every improvement you make is made against wrong data. Before you start, check three things:
- Is the right action being counted? For a form, measure a successful submission; for a purchase, the confirmation page after payment. A button click is not enough.
- Is the event marked as a key event? In GA4, the events that matter to your business are marked as key events. Marking only affects reports from the moment you mark it and does not change historic data. You first create or identify an event that measures the action, then mark that event.
- Is anything counted twice? Place a real test order and see how many times it appears in GA4, in the ad platform and in the order system.
The GA4 setup checklist takes you through the setup items in order; to see why GA4 and the ad platform show different numbers, read why GA4 and ad platforms disagree.
2. Don’t look at one average: split the rate
A site has a single conversion rate, but that number is the average of many different behaviours. A visitor who searched for your brand and one seeing you for the first time do not convert in the same way, and nor do desktop and phone. Put the rate side by side across these splits:
| Split | What to look at | What it tells you |
|---|---|---|
| Traffic source | Organic, paid, email, direct | Which channel’s visitors actually convert |
| Device | Desktop, phone, tablet | Whether mobile has a problem of its own |
| Landing page | The first page entered | Which page gets traffic but does not carry it to a sale |
| New and returning visitors | First visit and later visits | How many visits it takes to decide |
Usually one cell stands clearly apart from the others: say, phone conversion sits well below desktop while most of the traffic comes from phones. That is the place to work first; instead of trying to lift the average, you pick that cell.
3. Build the funnel and find where it leaks
Write the steps backwards from the target action. In e-commerce that is usually product page, add to cart, begin checkout and purchase; for an enquiry form it might be landing page, click on the form and successful submission. Funnel exploration in the Explore section of GA4 shows what share of visitors carries on at each step. A funnel can have at most 10 steps, but four or five are enough for a first pass. Because the steps rest on events, the measurement check in step 1 matters again here.
Two settings change the result. An open funnel lets visitors enter at any step, a closed one only at the first. And if you split by device, a visitor appears only under the device they entered the funnel with, even if they complete the later steps on another one. Keep that in mind when reading the gap between phone and desktop.
The step that loses the largest share on the way to the next is your work area. Many people adding to the cart but not starting checkout usually points to a problem on the payment page; few people adding to the cart at all points to missing information on the product page. Don’t compare with general averages; compare with your own earlier periods and the same split.
4. Find out why with behaviour data
The funnel tells you where you lose people, not why. For the why you need heatmaps and session recordings. A tool such as Microsoft Clarity shows where visitors click and scroll, and you can watch recordings one by one. According to Clarity’s FAQ, repeated rapid clicking in one area is flagged as a “rage click”, and a click that produces no response as a “dead click”. Both show a visitor expecting something to work that does not.
Recordings are kept only for a limited time; Clarity retains them for 30 days, and only favourites and a randomly selected sample are kept longer. So when you see a real problem, save a screenshot and a note straight away. Watching a few recordings from the step with the largest loss and writing down what they have in common gives you the first idea. This method is a source of hypotheses; the proof comes from the test.
5. Write a hypothesis
It is easy to turn a problem into an idea for a change, but an idea is not a hypothesis. A written hypothesis has four parts:
- Observation: What did you see, where, and in which data?
- Change: What will you do? It should be one change.
- Expected effect: Which behaviour do you expect to change, how, and why?
- Measure: Which number will you watch, and which guardrail number (returns or cancellations, say) will you keep an eye on while the test runs?
An example: “In the recordings we saw visitors leaving when the shipping cost appeared at checkout. If we move the shipping information to the product page, the share of people who start checkout and then buy will rise.” That can be tested; “let’s make the page feel more trustworthy” cannot.
6. Set up the A/B test and sample size properly
An A/B test shows two versions to visitors at random and asks whether a difference comes from chance or from the change. Decide each point before you start:
- Work out the sample in advance. The number of visitors needed depends on your current conversion rate, the smallest difference you want to detect and the confidence you want. The A/B test sample size calculator works it out at 95% confidence and 80% power and gives the duration in full weeks. Halving the difference you want to detect roughly quadruples the sample you need.
- Run in full weeks. Weekday and weekend visitors behave differently; a test that ends mid-week can mislead.
- Don’t stop because it looks good. Looking at interim results and stopping early raises the chance of declaring a winner where there is no real difference.
- Change nothing else during the test. A new campaign or a design update on the same page makes the result unreadable.
- Don’t harm search. Google prohibits showing one set of URLs to Googlebot and a different set to users (cloaking). For tests that use separate URLs it recommends rel=canonical pointing to the original, and for tests that redirect, a temporary (302) redirect instead of a 301.
Decide by asking this:
| Situation | Suggested route |
|---|---|
| The page has enough traffic and the difference can be tested with one change | A/B test |
| The sample calculation says months | Heatmaps, session recordings and user interviews; a bigger change |
| The problem is plainly a bug (broken button, form that does not work) | Fix it directly; don’t test it |
| The same problem appears on several pages | Test on the highest-traffic template and carry the result to the others |
What not to test
- A goal whose measurement is not verified. If the conversion event is wrong, the winner is wrong too.
- A page with no traffic. On a page with very few visitors, a fine difference cannot be picked up.
- Many things at once. If you change the headline, the image and the button together, you cannot know which one worked.
- Speed and bugs. A slow page or a broken form gets fixed without a test; we covered how speed affects sales in why a fast site sells more.
- Matters of taste. Colour and typeface preferences do not earn a place in the queue unless an observation backs them.
Common mistakes
- Going for the easiest change because the average rate is low, without looking for the real loss in the funnel.
- Running the test too short, or closing it the moment a winner appears.
- Not archiving the losing test; knowing what did not work makes the next test easier to choose.
- Raising the ad budget and not noticing the rate worsen; to look at the rate first, see measure before you raise the budget.
- Keeping results in scattered places. Each test’s hypothesis, duration, number of visitors and decision should sit in one place; the Monday report is a good home for reading chosen numbers weekly.
Common questions
What should you do first to increase conversion rate?
First confirm that the conversion event is counted correctly, then use funnel exploration to find the step with the largest loss. Any work started without these two checks rests on guesswork.
How much traffic does an A/B test need?
It depends on your current rate and the difference you want to detect. The lower the rate and the smaller the difference, the faster the number of visitors needed grows; find the figure with a sample calculation before the test starts. If the duration comes out in months, go with behaviour data and bigger changes instead of an A/B test.
Do you need a separate hypothesis for every page?
No, but you need one for every test. Start with the template that gets the most traffic and loses the most, and carry the result over to similar pages.
After reading this, there is one thing you can do today. Open a four-step funnel in the Explore section of GA4, view the latest period split by device, and write the name of the step with the largest loss on a piece of paper. Everything else starts from that step. If you would rather have an outside eye set up the measurement and the testing routine, we do this as part of our data and CRO service.
Sources
This article is general information, not legal, financial or technical advice. Platforms, their rules and the law change over time, so some details may have changed since 11 November 2026 or contain small errors. Before you make a decision for your business, check the current official sources or write to us.