E-Commerce Conversion Rates: 2026 Benchmarks by Industry

A.What the number is made of

E-commerce conversion rate is transactions divided by sessions, times one hundred. The arithmetic is trivial. The variance is not, and it decomposes along four dimensions that each carry more explanatory power than anything about your site design.

Industry is the largest. Price point, purchase complexity and repeat cadence set the ceiling before a designer touches anything.

The ordering tells you which lever is even available. Apparel pays for sizing uncertainty, and its apparent rate overstates net sales once 20% to 30% returns land. Electronics pays for cross-shopping: the research happens on your site and the purchase happens on whichever retailer wins the price comparison, which depresses the rate for every individual store regardless of quality. Luxury is the clean case of a category where a low rate is correct, because those sites exist for aspiration and for driving an in-store purchase.

Channel is the second dimension, and it is closer to a measure of intent than a ranking of marketing quality.

Email leads because the audience already raised its hand. Search converts in proportion to how precisely the landing page answers the query, and paid search sits slightly below organic only because it blends branded terms at 4% to 8% with non-branded at 1% to 2%, so a blended paid rate is mostly a statement about keyword mix. Direct is a mixture of returning customers and untracked traffic, which dilutes it. Paid social sits at the bottom because it interrupts people who were not shopping, and retargeting at 2% to 4% against prospecting at 0.3% to 0.8% is the same intent gradient appearing inside one channel.

Device is third and the most consistent pattern in the data: desktop 3.5% to 4.5%, tablet 2.5% to 3.5%, mobile 1.5% to 2.5%. Geography is fourth, running from the UK at 3.5% to 4.5% on compact delivery and mature returns infrastructure, down to markets where mobile-first shopping, cash on delivery and thinner logistics put the range at 1.0% to 2.0%.

One channel has no benchmark yet, and anyone quoting one is guessing. Shoppers increasingly start with a question to an AI assistant rather than a search box, arrive partway through evaluation because the assistant did the comparison, and carry no referrer, which drops them into your direct bucket. The LLM Acquisition report exists to separate those people from the crawlers the same assistants run, so you can measure your own rate for a quarter and use that as the benchmark.

B.So the headline moves when nothing has changed

Once you accept that the rate is a weighted average, the consequence follows immediately. Your blended conversion rate changes whenever the weights change, with no change whatsoever in how well the store sells. A quarter where paid social ran hard pulls the average down. A quarter where the email list grew pulls it up. Neither says anything about the site.

The device version of this is the one that catches teams out, because it can run backwards. If mobile converts at 2% and desktop at 4%, and your mobile share rises from 50% to 65% across a year, the blended rate falls from 3.0% to 2.7% even if both mobile and desktop improved over the same period. That is Simpson’s Paradox operating on the most reliable trend in e-commerce, and it means a declining headline is at least as likely to be a mix shift as a problem.

This is the first real constraint on benchmarking. Comparing your blended number to a published blended number compares two different weightings of the same underlying sub-rates, and the difference between them is mostly the difference in weights. The only comparison that survives is like for like: your mobile rate against a mobile figure, your paid social rate against a paid social figure, your category against your category. And the only comparison that is fully controlled is against yourself, in the same segment, in an earlier period.

A.A rate is an output, and outputs have no address

The conversion rate is the product of every stage rate between arrival and payment. Multiply four stage rates together and you get one number, which means the same number can be produced by wildly different failures. A store at 2.3% might be losing everyone at the product page and converting checkout beautifully, or the reverse, and the headline is identical in both cases.

Read the absolute losses rather than the percentages and the priority inverts against intuition. Checkout loses 150 people and gets all of the attention because 60% feels like a fixable number. The product page to cart step loses 5,250, and a single point of improvement there is worth more than perfecting everything downstream of it. Benchmarks are published at the level of the output, so they can tell you that 2.3% is low for your category and cannot tell you that the loss is concentrated in one step.

B.And the number you are comparing is partly a counting artefact

The second reason is that the denominator is a choice, and the conventional choice understates you. Almost every tool divides by sessions. A customer who visits three times and buys on the third converts at 33% by sessions and 100% by people. So a store with a long consideration cycle reports a lower rate than a store with an identical close rate and a shorter one, and reads it as underperformance.

Cross-device behaviour compounds it in a specific direction. Part of the mobile gap is not user experience at all: people research on a phone and buy on a laptop, and session-based tools record that as one failed mobile session and one successful desktop session. The mobile rate is understated, the desktop rate is overstated, and the conclusion a team draws from the gap, that the mobile experience is broken, may be partly an artefact of how the two visits were counted. Person-level tracking resolves the two visits to one buyer and shows the mobile session contributing to the sale it did not capture.

The most honest measure is at the person level: of the unique individuals who visited this month, what share bought? That is also the measure no published benchmark uses, which is the closing argument against benchmarking. The figure you can compare is the one you should not steer by, and the figure you should steer by has nothing to compare against. Attribution window and conversion definition add two more degrees of freedom on top, each of which produces a different rate from identical behaviour. Our guide to calculating conversion rate works through those choices.

A.Work the largest absolute loss, in the segment with the widest gap

Find the step that leaks and the segment it leaks in, then apply the fix that step actually calls for. Checkout friction is the best-evidenced of them: the Baymard Institute puts cart abandonment near 70%, with a long or complicated checkout named in 18% of cases. The lever there is structural rather than cosmetic. Every additional form field costs several percent of completions, forced account creation is one of the most cited abandonment reasons, and express payment methods remove the form entirely. Consolidating a four-page checkout into two is typically worth 10% to 35%, because every click is another chance to reconsider.

Speed is the second, and it is measured rather than argued: research from Google and Deloitte associates a one-second mobile improvement with up to 27% more conversions. Target Largest Contentful Paint under 2.5 seconds on mobile, and treat third-party scripts as the usual cause. Trust is the third, and it operates at the product page rather than at checkout: reviews, visible return policies, real customer photos, and shipping costs shown early instead of revealed at the last step. Personalisation is the fourth, worth 10% to 30% when the recommendations are genuinely useful, and it is the one with a data prerequisite. You cannot personalise for an anonymous cookie, so it depends on resolving visits to people before it depends on any recommendation logic.

None of that is a sequence to work through in order. The funnel decides which one you are allowed to care about this quarter, and the segment split decides who you are fixing it for. If mobile converts 60% below desktop after you have accounted for cross-device counting, a mobile-specific fix beats a site-wide redesign. That is what Funnels is for: a rate at every step, splittable on device, source, or any property you capture, so the loss has an address before anyone proposes a fix. For the step that most often pays, see our checkout optimization lessons and the broader funnel optimization guide.

B.Then change the target, because conversion rate can be won and lost at once

Conversion rate is a rate, and a rate is indifferent to the size of what converts. Any lever that raises the rate by lowering the value of an order can win the metric while losing money, and discounting is the standard way to do it.

Revenue per visitor is the better optimisation target, because it captures order value and conversion in one number and cannot be gamed by either alone. Run every test against it and the discount variant above fails on sight, without an argument about whether the conversion lift was real.

Revenue per visitor still has a horizon problem, though, which is the last thing to fix. It settles inside the session, and a customer acquired on a heavy promotion who buys once is worth a fraction of one from organic search who returns for years. The measure that closes that gap is lifetime value by acquisition source, first product bought, and region, which needs the same person key that fixed the denominator in II.B. The Revenue and Cohorts reports connect a first visit to repeat purchases months later, and Populations turns any behavioural definition of a valuable customer into a segment you can hold constant across reports. Our LTV walkthrough covers the arithmetic, and average order value strategies covers the other half of revenue per visitor.