Conversion Optimization: A Complete Guide

Baymard Institute keeps a running average of documented online shopping cart abandonment rates, compiled from dozens of separate studies. It has hovered around 70% for years (Baymard). Seven in ten shoppers who put something in a cart leave without buying it. Many of them were only comparing prices, and no design change will keep them. The rest left over something the store controls, and Baymard's checkout research estimates that fixing those problems could raise a large e-commerce site's conversion rate by about 35%. That is a relative lift: a store converting at 3% would reach about 4%.

What conversion optimization means

Conversion optimization (in the UK, conversion optimisation; in most job titles, conversion rate optimization or CRO) is the practice of increasing the share of visitors who complete an action you care about. The action is the conversion. On a store it's an order. On a SaaS site it might be a trial sign-up, a demo request, or an upgrade from the free plan. On a content site it might be a newsletter subscription.

The conversion rate is conversions divided by the visitors (or sessions) who had the chance to convert, over the same period. A pricing page that 8,000 people saw in March, producing 240 trial starts, converted at 3%.

Before measuring anything, settle which conversion counts. A page gets one primary conversion. If the sign-up page is judged on sign-ups, newsletter opt-ins and "contact sales" clicks at once, someone will report whichever one went up. Settle the denominator too: sessions, unique visitors, or visitors who reached a specific step. A rate computed over all sessions will drop the day a campaign brings in a wave of casual traffic, even if nothing on the page changed.

How a conversion funnel works

Most conversions are the last step of a sequence. A conversion funnel lists those steps and the number of people who reach each one. For a small online store over one month it might look like this:

Step	Sessions	Continue to next step
Product page	20,000	15%
Added to cart	3,000	50%
Started checkout	1,500	60%
Reached payment	900	two-thirds
Placed order	600

Overall, 600 orders from 20,000 product-page sessions is a 3% conversion rate.

The cart-to-checkout step loses half of the people who reached it, which makes it look like the obvious target. Run the arithmetic first, because a funnel multiplies. Raise cart-to-checkout from 50% to 60%, a 20% relative lift, and if the later steps hold their rates the result is 1,800 checkouts, 1,080 at payment and 720 orders. Raise product-to-cart from 15% to 18%, also a 20% relative lift, and you get 3,600 carts, 1,800 checkouts, 1,080 at payment and the same 720 orders.

So the weakest-looking step is not automatically the one worth fixing. In practice the later rates don't hold perfectly (extra shoppers pulled into the cart are often less committed than the ones already there), but the arithmetic moves the question to where a lift of that size is easiest to get and to prove. On this store the two tests would take about the same time: the product page has nearly seven times the cart's traffic, but its 15% baseline needs a much larger sample to detect the same relative lift.

Finding where and why people leave

Ronny Kohavi and Stefan Thomke report that at Microsoft only about a third of experiments improved the metric they targeted (HBR). Research before the test is how a team does better than that, and analytics alone won't supply it, because it shows the step where people leave and nothing about the reason.

Start with the segmented funnel. Break it down by device, traffic source, new versus returning visitor, and country. A checkout that converts at 70% on desktop and 40% on mobile is two problems, and the average hides both. Build these views once and keep them; the guide to analytics dashboard design covers how to make them readable.

Once the funnel names a step, go to the behaviour data for that step. Filter session recordings to people who left there; twenty of those will teach you more than two hundred random visits, and the rage clicks and back-and-forth between pages show you where they got confused. A scroll heatmap tells you whether the shipping cost or the main call to action sits below the point most visitors reach, and a click heatmap shows clicks on things that aren't links. If the step is a form, field-level analytics shows which field people abandon on and which ones they correct most. A required phone number with no reason given is a common culprit.

Then ask people. Five to eight participants attempting the purchase while thinking aloud will surface problems click data can't explain, such as a shopper who can't tell whether the date shown is dispatch or delivery; the usability testing guide covers running the sessions. A one-question exit survey ("What stopped you from completing your order today?") and the support team's most common tickets give you the customer's own wording. A heuristic evaluation of the step is the cheapest of all, though it can't say how many people a problem affects, so treat its findings as candidates for the other methods to confirm.

Turning findings into tests

Each finding becomes a hypothesis that names the change, the expected effect on the primary conversion, and the evidence behind it: "Show the delivery cost on the product page, which should raise cart-to-checkout, because exit surveys and recordings show people leaving the cart when shipping appears." The A/B testing guide covers how to write one and the rules for sample size, significance and stopping.

Rank the list before testing anything. ICE scores each idea on impact, confidence and ease; PIE uses potential, importance (how valuable the page's traffic is) and ease. Neither score is precise. They make the team say out loud why one idea goes before another, and ICE's confidence column is where the research from the previous section earns its place.

Low-traffic sites can't test everything. If a checkout gets 300 orders a month, a test hunting a small lift may need several months to reach a reliable result, and an experimentation platform the site can't feed with traffic is wasted money. Fix the problems research found with high confidence (a broken field on mobile, a hidden shipping cost) and ship them without a test. Save testing for the changes you are actually unsure about, and put the budget into recordings, interviews and usability sessions instead.

Judge each test on the conversion it was meant to move, and keep watching after it ends. A change that raises trial sign-ups but brings in people who never activate hasn't helped. Compare the variant's later behaviour, such as activation, refunds and second purchases, against a proper control group before calling the result a win.

Where most conversion gains come from

Across studies of checkout and sign-up flows, the same problems keep appearing. Use them as places to look during research.

Cost. In Baymard's abandonment surveys, extra costs that are too high (shipping, taxes, fees) is the reason shoppers give most often, well ahead of anything about the interface. Being unable to see or calculate the total before checkout is a separate reason on the same list. A store can't always lower its shipping price, but it can stop the total from being a surprise: show it, or a reliable estimate, on the product page and in the cart. That tends to cost a little at the cart and recover more at payment, which is the interaction between steps the funnel arithmetic above assumes away.

Forced account creation. Requiring an account before purchase is also high on Baymard's list. Guest checkout, with the option to save details after the order is placed, keeps the sale.

Long forms. Remove fields the business doesn't use, fill city from postcode where your data allows, and mark optional fields clearly. Baymard's benchmarks find most checkouts ask for more fields than the order needs.

Payment doubt and slow pages are smaller but cheap to check: a card form that looks different from the rest of the site, or offers none of the payment methods your customers already use, makes people hesitate, and a slow load on mobile loses people before they see the page at all.

On landing pages for sign-ups and lead forms, the leak usually sits higher up. The visitor doesn't understand what they're getting, or what it costs, quickly enough to bother. A headline that names the outcome, a visible price or a clear "free" with its limits, and one primary call to action per page leave the visitor less to work out.

Dark-pattern tactics

Countdown timers that reset on reload, fake low-stock warnings, pre-ticked add-ons and cancellation flows that hide the cancel button can all raise a conversion rate in the short term. Their cost turns up later as refunds and churn. The retention strategies article covers how that happens and offers a disclosure test for telling these tactics apart from honest ones; showing shipping costs earlier passes it, and a fake timer doesn't.

Conversion optimization FAQ

What is conversion rate optimisation?

Conversion rate optimisation is the British spelling of conversion rate optimization: the practice of raising the percentage of visitors who complete a target action, such as a purchase or a sign-up, by finding and fixing what stops them.

Is conversion optimization the same as CRO?

Yes. Conversion optimization, conversion rate optimization and CRO describe the same work. Some teams use "CRO" for the experimentation program specifically and "conversion optimization" for the wider effort, including research and fixes shipped without a test.

What is a good conversion rate?

There isn't a single benchmark worth aiming at. Rates differ by industry, price, traffic source and how the conversion is defined, and published averages mix all of those together. Compare your rate to its own history, by segment, and judge changes against that.

How is conversion optimization different from SEO?

SEO brings more of the right visitors to the site. Conversion optimization raises the share of those visitors who act.

How do you reduce cart abandonment?

Find out why your shoppers leave, using exit surveys, recordings of abandoned checkouts and a usability test. Then check the common causes: high or hidden costs, forced account creation, long forms, missing payment methods and slow pages. Fix the confirmed ones first.

What tools does CRO need?

Analytics for funnels and segments, a behaviour tool for heatmaps, recordings and form analytics, a research tool for surveys and remote usability tests, and an experimentation platform once traffic is high enough to use one. Many products cover more than one of these. Whatever behaviour tool you install, mask form fields and personal data by default and check that it respects your consent banner.

How long does conversion rate optimization take to show results?

Fixes for clear problems found in research can show up within weeks of shipping. A testing program takes longer, because each test needs enough traffic to give a reliable answer, and on low-traffic sites a single test can run for several months.

More on research methods, analytics and product strategy: the product insights hub.

Sources: Baymard Institute — Cart Abandonment Rate Statistics · Baymard Institute — Checkout Usability research · Kohavi, R. & Thomke, S. (2017), The Surprising Power of Online Experiments, Harvard Business Review.