Search how to increase conversion rate ecommerce stores see and you get lists. Nineteen tactics, twenty-five hacks, thirteen ways to do it right now. Most of the items are true somewhere, and not one of the lists tells you which of them is true for your store, which was the only thing you needed to know.
This page does two things instead of listing. It puts the work in order, because these changes interact and doing them in the wrong sequence means doing several of them twice. And it names a publisher in the same sentence as every number, because no corner of marketing carries more famous statistics with no readable source behind them than this one.
Said before the advice rather than after. We build and rebuild online stores for a living, so a page from us explaining that your store is the problem is interested testimony. Everything below is something you can do inside the admin you already have. The last section says what we would look at first and what to ask anybody who offers to do it for you, including us.
What the number is, and the three ways it gets computed wrong
The formula is not the hard part. Orders divided by sessions, times one hundred. The hard part is that nobody agrees what goes on the bottom.
Three denominators are in common use and they are not close to each other. Sessions counts every visit, so one person who comes back four times before buying drags the rate down. Users counts people, so that same shopper is one, and the rate goes up. Carts created counts only shoppers who got as far as adding something, so the rate jumps again and now measures your checkout rather than your store. All three are defensible. Reporting one while your agency reports another is how two people argue for a month about a number they both calculated correctly.
Two more things quietly move it. Most analytics cannot tell that the phone at lunchtime and the laptop at night are one shopper, so any store where people research on one device and buy on another is splitting its own customers in two. And unfiltered bot and internal traffic inflates the denominator with visits that were never going to buy, which makes every improvement you make look smaller than it was.
So before any of the fixes below, write down which denominator you use and stop changing it. A store that switches definitions halfway through cannot tell a real gain from a redefinition, and our walkthrough of what analytics can and cannot tell you covers where the rest of these gaps come from.
Where this page stops, and what sits next to it
Two boundaries, because both neighbors get confused with this and for different reasons.
The first is method. This page is the store-specific how-to, what to change on a catalog, a product page and a checkout and in what order, while our company website analysis methodology is the general procedure for judging any site’s experience and for testing whether a statistic you have been handed has a source. If you want the filter, read that one. If you want the store work, stay here.
The second is search. People ask what the difference is between SEO and conversion work, and the clean version is that one decides how many sessions arrive and the other decides what those sessions do. They are not rivals, but they do compete for the same pixels, and a fix for one can cost the other. Thinning a category page’s copy to get the grid higher can cost you the rankings that filled the page, which is why our guide to SEO for ecommerce and this page should be read together rather than one after the other.
One practical consequence. If your traffic is small, conversion work has a ceiling you will hit quickly, because a percentage of very little is still very little. Fixing a checkout that loses four in ten decided buyers is worth doing at any traffic level. Running tests to find a gain of a tenth of a percent is not.
The figures in this field, and which ones have a publisher
This subject runs on repeated numbers, so it is worth separating the ones you can check from the ones everybody quotes and nobody has read.
The best-sourced body of work here is Baymard Institute’s, because it shows its working. Baymard publishes an average documented cart abandonment rate of 70.22 percent and states on the same page that the value is calculated from fifty different studies, then lists all fifty with the publisher and the date each was retrieved. The list was last updated on 22 September 2025. The useful part is not the average. It is that the individual studies in that list run from 55.00 percent, Forrester Research in 2010, to 84.27 percent, SaleCycle in 2020. A spread of nearly thirty points is telling you that abandonment depends enormously on what you sell and to whom, which is exactly what a single headline average hides.
Now the other kind. Two numbers appear in almost every article on this subject and neither survives being followed home. The claim that every hundred milliseconds of added load time cost Amazon one percent in sales traces back to remarks made at a conference in 2006 and to no published study anybody can open. The claim that a one second delay cuts conversions by seven percent is attributed to an Aberdeen Group report from 2008 that is not publicly readable, so every citation of it is a citation of somebody else’s citation. Both may well be directionally right. Neither is evidence, and a store owner who reorders a roadmap around them has reordered it around folklore.
The rule this page follows, and the one worth applying to anything else you read, is that a figure needs a publisher named beside it and a page you can open. Our piece on what published benchmarks are actually good for goes further into why two publishers’ numbers can never be averaged into a third.
How to increase conversion rate ecommerce stores can measure
Four passes, in this order, and the order is the part that gets skipped.
First the checkout, because everybody in it has already decided to buy from you and losing them is the most expensive failure available. Second the product page, because that is where the decision is made or abandoned. Third speed, on mobile specifically. Fourth, and only fourth, testing.
The order is not a preference. It is forced by the fact that each pass corrupts the measurement of the ones below it. If your checkout drops a fifth of the people who reach it, then every product page improvement you make is being measured through a leak, and a real gain on the page will show up as a smaller gain in orders. Fix the leak and the same page change suddenly looks better, which tempts you to credit the wrong work. The same applies downward. Speed changes move which sessions reach the checkout at all, so a test run across a speed fix is a test with a moving population.
Work one pass at a time, let each one settle for long enough to see whole purchase cycles, and write down what you changed and when. A store that cannot say what changed in which week cannot attribute anything afterward, and attribution is the entire reason for doing the work in an order.
The checkout pass, where the published evidence actually is
This is the one pass where you do not have to guess, because somebody has asked shoppers directly and published what they said.
Baymard’s page on the subject reports that “our latest quantitative study of reasons for abandonment found that 42% of US online shoppers have abandoned a cart” because they were only browsing, which it treats as largely unavoidable. Set that group aside and Baymard’s published distribution of the remaining reasons runs, in its own order, 40 percent extra costs too high, 20 percent delivery too slow, 19 percent not trusting the site with card details, 18 percent being made to create an account, 17 percent a checkout that is too long or complicated, 17 percent site errors, 13 percent an unsatisfactory returns policy and 12 percent not being able to see the total cost up front.
Read that list as a work order and four of the eight are design decisions already in your hands. Show shipping, tax and fees on the cart page rather than at the final step, which kills the top reason and the eighth one together. Offer a guest checkout, which removes the fourth. Count the fields in your form and delete every one you do not use, which shortens the fifth. None of that needs a rebuild and none of it needs a test to justify, because the failure being fixed is a documented reason people gave for leaving.
On scale of the opportunity, Baymard’s cart and checkout benchmark scores 65 percent of the sites in it as mediocre or worse, 35 percent as decent or better and only 2 percent as good, across a benchmark it describes as 344 top-grossing US and EU ecommerce sites. It states that “The average site has 32 unique improvements to perform in their checkout flow” and puts the potential conversion gain for a large-scale store from better checkout experience at 35 percent, from its own usability test sessions. Treat the 35 percent as a ceiling somebody measured on large sites rather than as a promise about yours, and treat the 32 improvements as the more useful half, because it tells you this is a list of small defects rather than one big idea.
The product page pass, and the questions it has to answer
A product page has one job, which is to answer every question standing between a shopper and a decision, in the order they occur to a person who cannot pick the thing up.
There are four of those questions on almost every store. Will it fit or suit me, which is sizing, dimensions and scale, and a photograph of the object next to something recognizable beats any table. What does it cost delivered, which is the same reason the checkout list puts extra costs first. When does it arrive, stated as a date rather than a shipping class nobody outside your warehouse understands. And what happens if it is wrong, which is your returns policy, sitting on the page rather than in a footer link, since 13 percent of abandoning shoppers in Baymard’s list named the returns policy directly.
Trust belongs on this page too, and here the evidence is specific. Nineteen percent of abandoning shoppers in that same list said they did not trust the site with their card information. That is a design problem with design answers, which are a real business address, a visible returns and contact route, product photography that is obviously yours rather than the manufacturer’s stock set, and reviews that include the unflattering ones. A page carrying only five-star reviews reads as curated, and curated reads as untrustworthy.
Two things worth saying plainly about trust, because both get sold hard. Trust badges are a graphic, and a graphic is not evidence of anything a shopper can verify. And people ask whether a domain ending in .co converts worse than one ending in .com. We could not find a controlled study either way, and the pages arguing about it are published by companies that sell domain names. The mechanism that does have evidence behind it is the one above. Shoppers leave when they cannot verify who you are, and that is fixable on the page whatever your domain says.
The speed pass, on the device most of your traffic is using
Speed earns its place in the order because it decides how many people reach the pages you just fixed, and on phones it is usually much worse than the owner thinks.
Two published sources are worth having here and both put a named publisher behind the number. Google states on its own AdSense help page about mobile page speed that “Our research shows that 53% of visits are likely to be abandoned if pages take longer than 3 seconds to load.” Note what that claim is and is not. Google is naming its own research on its own domain, which is a publisher standing behind a number, but the link it puts on that figure now lands on a general marketing resources page rather than on the study, so you can cite Google for the claim and not for the method. Separately, the agency Portent published an analysis in April 2022 across twenty sites, six of them B2C ecommerce, using page speed data from 5.6 million sessions in a thirty day window, and reported that a site loading in one second had an ecommerce conversion rate 2.5 times higher than one loading in five seconds. Portent also reported that the same comparison against ten second loads came out lower, at 1.5 times, and said in the piece that this may be down to its smaller sample in that range. A source that publishes its own awkward result is worth more than one that does not.
For a target rather than a scare, use the published threshold instead. Google’s Web Vitals documentation states that “LCP should occur within 2.5 seconds of when the page first starts loading” and measures that at the 75th percentile of real loads rather than in a lab. On a product page the element that decides Largest Contentful Paint is nearly always the main product image, which means the fix is usually not a plugin. It is serving that one image at the size it displays, in a modern format, without a carousel script deciding when it is allowed to appear.
Measure on field data from real visits, not on a score from your own laptop. Your laptop is on your office connection with your browser cache warm, which is the one visitor you do not need to worry about.
Why most store tests never finish, and what to do instead
Testing is the fourth pass and for most stores it is the one that should be skipped, which is unpopular advice and is arithmetic rather than opinion.
An A/B test detects a difference by accumulating enough events to rule out chance. When the thing you are counting happens to a small percentage of visitors, you need a great many visitors before a modest relative improvement becomes distinguishable from noise, and the smaller the improvement you are hunting the more you need. That is why a store taking a few hundred orders a month can run a test for a two weeks, watch one variant lead the whole time, and still have learned nothing that will survive next month.
What happens next is the real damage. Somebody calls the test, ships the winner, the number drifts back, and the organization concludes that conversion work does not do anything. It did nothing because it was never measurable at that volume, not because the change was wrong.
So if you are not at the traffic level that supports testing, fix known defects instead and accept before-and-after measurement with its limits stated. A guest checkout option, a delivery date on the product page and a cart that shows the total do not need proving in your specific case, because they remove reasons shoppers have already stated for leaving. Save testing for the questions where reasonable people disagree and where you have the volume to settle it. Our guide to running the audit that finds those defects is the pass that produces the list.
What we would not touch first
The fastest way to spend a quarter and learn nothing is to start with the things that feel like progress.
A full redesign is top of that list. It changes everything at once, which means that whatever happens to the number afterward, you cannot say which change caused it, and it usually resets the small accumulated fixes that were working. Theme swaps have the same problem in a cheaper package. Popups and urgency timers are next, not because they never lift a number in the short term but because they lift it by pressuring the same shoppers whose stated reason for leaving was that they did not trust you.
Then there is the habit underneath all three, which is changing five things in one week. It feels efficient and it destroys the only asset this work produces, which is knowing what worked. One pass, one week, one written record.
Last, do not start by comparing yourself to an industry average. Every publisher of those tables computes them from a different panel with a different definition, and the gap between two of them is larger than most improvements you will ever make. If you sell in a category with a long consideration cycle, your rate should be lower than a table says and that is information rather than failure. Our write-up of auditing a store at catalog scale and our look at retail SEO both deal with the same trap from the traffic side.
Where we stand, since we build the things this page is about
We design and rebuild ecommerce stores, which means rebuild it is the conclusion we profit from, and you should read the section above about redesigns knowing that we wrote it against our own interest.
Given a store we had never seen, we would open the checkout on a phone and try to buy something, counting the steps and the form fields and noting where the total first becomes visible. Then the product page for the best seller, against the four questions above. Then field speed data on that product page, not a lab score. That takes under an hour and it produces a list of defects with reasons attached, which is a different object from a list of tactics.
If somebody is quoting you for this work, three questions sort the field quickly. Ask which denominator they will report the rate on, and whether it is the one you already use. Ask for the source of any statistic in their pitch, and watch whether a publisher and a page come back or a confident restatement does. And ask what they would do first and why that order, because an answer that starts with a redesign is an answer about their capacity rather than your store. We sell ecommerce website design, so put those three questions to us as hard as to anyone.
You do not need any of that to start. The checkout pass is free, it is an afternoon, and the reasons people gave for leaving are already published. If you would rather have the list written up first, our free website audit is where that starts.
