Average Ecommerce Conversion Rate by Industry (2026): Every Benchmark, With Its Source

The average ecommerce conversion rate in 2026 sits between 1.4% and 2.74%, depending on which dataset you trust. Dynamic Yield's Mastercard-owned benchmark puts the global average at 2.74% (Dynamic Yield, 12-month rolling average, retrieved July 2026), Statista counted 1.6% of global ecommerce visits converting in Q3 2025 (Statista, as cited by Shopify), and Littledata's panel of 2,800 Shopify stores averaged 1.4% (Littledata, 2023).

Those three numbers are not in conflict. They measure different stores, in different years, with different denominators. That spread, and what it means for your store specifically, is what this post unpacks, one sourced row at a time.

The short answer: what “average” means in 2026

TL;DR Published averages range from 1.4% to 2.74% because different panels measure different stores. Pick the row closest to your situation, note its vintage, and treat it as context, not a target.

Ask five benchmark reports for the average ecommerce conversion rate and you will get five defensible numbers. Here is the full set this post relies on, in one table. Every later section references a row from it.

Figure	Source	Sample	Vintage	Caveat
2.74% global average conversion rate	Dynamic Yield (Mastercard)	Customer sites, size undisclosed	12-mo rolling, retrieved Jul 2026	Enterprise-skewed panel; changes monthly
1.6% of global ecommerce visits converted	Statista, as cited by Shopify	Global ecommerce visits	Q3 2025	Primary is paywalled; cited exactly as Shopify does
1.4% Shopify-store average (mobile 1.2%, desktop 1.9%; top 20% above 3.2%, top 10% above 4.7%)	Littledata	2,800 Shopify stores	2023	Client base skews smaller stores; older vintage
Industry benchmarks from 0.94% (luxury) to 6.22% (food & beverage)	Shopify blog	Undisclosed	Past 12 months, published 2026	No sample size or methodology disclosed
Regions: EMEA 2.89%, Americas 2.69%, APAC 1.58%	Dynamic Yield	Same rolling panel	Retrieved Jul 2026	Same caveats as above
Average order value: global $185; desktop $255 vs mobile $164; luxury & jewelry $386	Dynamic Yield AOV benchmark	Same rolling panel	Retrieved Jul 2026	Per-order value; same caveats as above
Mobile 51.51% of worldwide web traffic (desktop 47.12%)	Statcounter Global Stats	Global web traffic, all site types	June 2026	All web traffic, not ecommerce-specific; changes monthly
70.22% average cart abandonment rate	Baymard Institute	Average of 50 studies	Retrieved Jul 2026	Meta-average across mixed methodologies
35.26% modeled conversion gain from better checkout; $260B recoverable	Baymard Institute	Checkout-usability research, US + EU	Retrieved Jul 2026	Modeled potential, not a measured A/B result
A 1-second site converts at 2.5x the rate of a 5-second site	Portent	20 sites, of which 6 ecommerce	2022 update	Small sample; correlational
0.1s faster mobile load → retail conversions +8.4%	Deloitte, “Milliseconds Make Millions”	Retail, travel, luxury brands, EU + US	2020	Commissioned by Google
40% of visits showed frustration signals; conversion down 6.1% YoY	Contentsquare, 2025 benchmark	90 billion sessions, 6,000 sites	Q4 2023 vs Q4 2024	Proprietary “frustration” definition; correlational
Conversion down another 5.1% YoY; average order value up 6%	Contentsquare, 2026 benchmark	99 billion sessions, 6,500+ sites	2026 edition, retrieved Jul 2026	Headline figures from the ungated summary; full report gated
Median org: ~10% of experiments move their target metric (Microsoft 33%, Bing 15%, Airbnb Search 8%)	Ronny Kohavi, GrowthBook interview	Large experimentation orgs	Retrieved Jul 2026	Mature, heavily optimized products
33.5% of 1,001 real A/B tests were significant winners; median lift across all tests 0.08%	Georgiev, Analytics-Toolkit	1,001 A/B tests	2022	Expert-practitioner panel; win rate biased upward
~70% of 115 public A/B tests were underpowered; median detectable effect 20.84% at 90% power	Georgiev, Analytics-Toolkit	115 tests from GoodUI's public database	2018	Public, self-selected tests; author recomputed the stats
165 winning tests out of 633 (~26% significant winners)	GoodUI evidence base	633 tests, 147M visitors	Retrieved Jul 2026	Self-selected contributed tests; counts grow monthly
Add-to-cart lift predicts order lift at R = 0.4983; checkout visits at R = 0.6085	GoodUI correlation study	44 to 533 experiments per correlation	Retrieved Jul 2026	Observational meta-analysis
One Bing headline test: revenue +12%, over $100M/year	Kohavi & Thomke, HBR	Single experiment	September 2017	An outlier, cited as the argument for testing everything

Every figure used anywhere in this post, in one place. Rolling sources (Dynamic Yield, Baymard, GoodUI) were re-fetched on July 14, 2026 and will drift after publication; the source link on each row leads to the live number.

Three things follow from this table, and they shape everything below.

First, the range is real and it is explainable. Littledata's 1.4% comes from 2,800 mostly smaller Shopify stores measured in 2023. Dynamic Yield's 2.74% comes from an enterprise-heavy customer base on a rolling 12-month window. Statista's 1.6% counts visits, globally, in a single quarter. Different populations, different clocks, different denominators. Any page that quotes one of these as “the” average without naming the panel is quoting a report it has not read.

Second, the number you should care about least is the global average. The industry table in the next section spans 0.94% to 6.22% within a single source. Your vertical explains more of your conversion rate than most tactics ever will.

Third, the only benchmark that pays your bills is your own trailing baseline. A store that moved from 1.1% to 1.3% made real money. A store sitting at 2.0% and feeling smug about “beating the average” may be leaving half its potential on the table if its category median is 3.9%. The rest of this post exists to help you find the right row, and then beat your own number, not somebody else's. If you sell on Shopify, we keep a separate, deeper set of sourced Shopify-specific benchmarks, and the tactics side lives in the complete Shopify CRO playbook.

Average ecommerce conversion rate by industry

TL;DR Within Shopify's 2026 category benchmarks, industry conversion rates span 6.6x, from food and beverage at 6.22% down to luxury and jewelry at 0.94%. Compare yourself to your own row and nobody else's.

Two named panels publish per-industry conversion rates that we fetched and read for this post: Shopify's category benchmarks covering the past 12 months (Shopify blog, 2026; sample size undisclosed, and the article itself warns against treating benchmarks as universal) and Dynamic Yield's live benchmark page (12-month rolling, retrieved July 2026; sample size undisclosed, enterprise-skewed). Here they are side by side. Where a cell is empty, that panel did not publish a figure we verified at retrieval, and we will not fill the gap with a guess.

Industry	Shopify (past 12 months, 2026)	Dynamic Yield (retrieved Jul 2026)
Food & beverage	6.22%	–
Beauty & personal care	4.94%	5.37%
Multi-brand retail	3.93%	–
Fashion & apparel	3.06%	–
Home & furniture	1.41%	–
Luxury & jewelry	0.94%	0.71%

Shopify blog and Dynamic Yield benchmark page, both retrieved July 2026; samples and windows differ, see the methodology section. Empty cells mean that panel published no figure we verified.

Source: Shopify blog category benchmarks, past 12 months, retrieved July 2026. Sample size undisclosed. The coral bar marks the low outlier, which is also the highest-AOV category in the set.

The span inside one panel is 6.6x: food and beverage at 6.22% against luxury and jewelry at 0.94%, both from the same Shopify dataset. Read that again before you compare your store to any blended average. A supplements brand and a fine-jewelry brand could run identical sites, identical checkout flows, identical ad creative, and still land five percentage points apart, because purchase deliberation, price point, and reorder behavior differ that much by category. Cross-industry comparison is close to meaningless. Cross-panel comparison within an industry tells you something too: the two panels disagree on beauty by about 0.4 points and on luxury by about 0.2 points, and where panels disagree, panel composition is doing the talking, not the industry.

Here is each row on its own terms.

Food & beverage: 6.22%

Food and beverage leads Shopify's 2026 category benchmarks at 6.22% (Shopify blog, past 12 months, sample undisclosed). The mechanics favor it: low price points, consumable products that generate repeat orders, and buyers who already know what the product tastes like on visit two. A returning-customer purchase converts differently from a cold first visit, and food skews heavily toward the former. If you sell coffee or snacks and convert at 3%, the blended ecommerce average is flattering you while your own category median says otherwise.

Beauty & personal care: 4.94% to 5.37%

Beauty is the one category where both of our verified panels publish a figure: 4.94% in Shopify's 2026 benchmarks and 5.37% in Dynamic Yield's rolling panel (retrieved July 2026). The 0.4-point gap between them is a live example of panel composition at work; Dynamic Yield's enterprise-skewed base likely includes more established brands with strong repeat purchasing. Beauty shares food's replenishment dynamics, with the added weight of shade-matching and ingredient anxiety on first purchase, which is why product page work tends to matter disproportionately here.

Multi-brand retail: 3.93%

Multi-brand retailers convert at 3.93% in Shopify's 2026 benchmarks (Shopify blog, sample undisclosed). Selection breadth cuts both ways: more products match more intents, but comparison shopping inside a large catalog stretches sessions and splits attention across items. A multi-brand store's conversion rate is also the hardest to interpret, because its category mix is itself a blend. A retailer weighted toward consumables will read closer to the food row; one weighted toward considered purchases will read closer to home and furniture.

Fashion & apparel: 3.06%

Fashion and apparel sits at 3.06% in Shopify's 2026 benchmarks. It is the category most punished by uncertainty a screen cannot resolve: fit, fabric, and how the thing actually drapes. That is why returns run high and why the conversion battle in apparel happens on the product page, in size guidance, photography, and reviews rather than in the cart. If you run an apparel store, product page optimization is your highest-density testing ground, and 3.06% is your reference line, not 2.74%.

Home & furniture: 1.41%

Home and furniture converts at 1.41% (Shopify blog, 2026), a number that looks alarming next to beauty and is completely normal for the category. High prices, long deliberation cycles, delivery logistics, and purchases that often involve two decision-makers all compress conversion. A furniture store's visit-to-purchase window can span weeks, which means a per-session conversion rate structurally undercounts eventual buyers. If this is your row, session-level benchmarks deserve extra skepticism and your email capture rate may matter as much as your conversion rate.

Luxury & jewelry: 0.94% (or 0.71%)

Luxury and jewelry anchors the bottom of both panels: 0.94% in Shopify's 2026 benchmarks and 0.71% in Dynamic Yield's rolling figures (retrieved July 2026). Neither number is a problem to fix. High average order values mean a luxury store can out-earn a 5%-converting snack brand on a fraction of the orders; revenue per visitor, not conversion rate, is the honest scoreboard here, and raising average order value is often the better lever than chasing conversion points. A jewelry store converting at 1.3% is beating its category by roughly 40%, whatever the global average says.

One caution before you bookmark this table. Shopify does not disclose its sample size or methodology on the page, and cautions its own readers against universal benchmarks. Dynamic Yield's figures change monthly, which is why every citation above carries “retrieved July 2026.” Both facts get a full autopsy in the methodology section below. Use these rows as orientation, not as targets.

Conversion rate by device, region, and platform

TL;DR Desktop converts at 1.9% vs 1.2% on mobile (Littledata, 2,800 Shopify stores, 2023), EMEA leads regions at 2.89% (Dynamic Yield, retrieved July 2026), and the Shopify-panel vs all-ecommerce gap is a composition effect, not a platform verdict.

Industry is the biggest slicer, but three more cuts change what “average” means for you: the device your customers hold, the region they buy from, and the panel your benchmark was measured on.

Cut	Segment	Conversion rate	Source
Device	Desktop	1.9%	Littledata, 2,800 Shopify stores, 2023
Device	Mobile	1.2%	Littledata, same panel, 2023
Device	Blended average	1.4%	Littledata, same panel, 2023
Region	EMEA	2.89%	Dynamic Yield, rolling 12-month, retrieved Jul 2026
Region	Americas	2.69%	Dynamic Yield, retrieved Jul 2026
Region	APAC	1.58%	Dynamic Yield, retrieved Jul 2026
Platform panel	Shopify stores	1.4%	Littledata, 2,800 stores, 2023
Platform panel	All ecommerce	2.74%	Dynamic Yield, rolling, retrieved Jul 2026

The two “platform” rows are not directly comparable: different panels, different vintages, different store-size mixes. That is the point of the platform subsection below.

The desktop–mobile gap is the most actionable row in this post

In Littledata's panel, desktop sessions converted at 1.9% and mobile sessions at 1.2% (2,800 Shopify stores, 2023). Desktop buyers converted at roughly 1.6 times the mobile rate.

Why this row matters more than the others: blended averages hide it. Run the arithmetic on a store with 70% mobile traffic at exactly Littledata's segment rates. Seventy percent of sessions converting at 1.2% plus thirty percent converting at 1.9% blends to 1.41%, almost exactly the panel's overall average. That store's dashboard shows a “normal” conversion rate while the segment view shows most of its traffic converting at the panel's weakest rate. The blended number reports that everything is fine. The device split reports where the money is leaking.

So before you benchmark your store against anything in this post, split by device first. If your mobile rate is at or above your desktop rate, you are unusual and probably doing something right. If your mobile rate is less than 60% of desktop, worse than the panel's own ratio, you have a device problem wearing an average's clothing. We cover the specifics in our guide to mobile conversion on Shopify, and one preview from the evidence table: Deloitte measured a 0.1-second mobile speed improvement lifting retail conversions by 8.4% (Deloitte, “Milliseconds Make Millions,” 2020; Google-commissioned, which we flag every time we cite it).

One vintage warning, stated here rather than buried: the Littledata figures are from 2023. They are the most recent per-device Shopify-panel numbers we could verify at a named source, and they are two to three years old. Treat the ratio as more durable than the absolute values.

Region: a 1.3-point spread across the world

Dynamic Yield's rolling panel splits its 2.74% global average into EMEA at 2.89%, the Americas at 2.69%, and APAC at 1.58% (retrieved July 2026; sample undisclosed, enterprise-skewed). The EMEA–Americas gap is small enough to ignore for most decisions. The APAC gap is not, and it is a composition story as much as a behavior story: different marketplace dynamics, different payment norms, and a different mix of sites inside the panel itself.

The practical use of the regional row is narrow but real. If you ship internationally and your APAC sessions convert at half your domestic rate, that is roughly what the panel would predict, and it may be a currency, payment-method, or shipping-cost artifact rather than a site defect. Check what the panel says before declaring an emergency, and check your own segment data before trusting the panel.

The platform question: why the Shopify panel reads lower than “all ecommerce”

Put the two panels next to each other and an ugly-looking gap appears: Shopify stores at 1.4% (Littledata, 2023) against a 2.74% all-ecommerce average (Dynamic Yield, retrieved July 2026). More than one page on the internet has turned that gap into “Shopify stores convert worse than average.”

That reading does not survive contact with the methodology. Littledata's panel is 2,800 Shopify stores, and Littledata itself notes its client base skews toward smaller stores. Dynamic Yield's panel is its own customer base, which skews enterprise; companies that buy an enterprise personalization platform are, almost by definition, large operations with mature funnels. Small stores convert lower than enterprises on any platform. Measure a small-store panel and an enterprise panel and the small-store panel will read lower every single time, regardless of what software either group runs on. The 1.4%-vs-2.74% gap is a composition effect, not a platform verdict. Add the vintage mismatch, 2023 fixed-window data against a rolling 2026 window, and the comparison gets even less legitimate.

What you can honestly take from the platform rows: if you run a small or mid-sized Shopify store, Littledata's 1.4% median and its percentile ladder are your most representative published reference, precisely because of the small-store skew that makes it a bad stick for measuring the platform. If you run an enterprise store, Dynamic Yield's panel is closer to your peer group. The full Shopify-specific picture, including how benchmarks shift by traffic source and price point, lives in our sourced Shopify-specific benchmarks.

Which raises the question the next section answers: if the average depends this much on the panel, what number should you actually be aiming for?

What is a good conversion rate for ecommerce? Percentiles, not averages

TL;DR “Good” is a percentile inside your vertical and device mix, never a universal number. Across 2,800 Shopify stores, the median converted at 1.4%, the top 20% above 3.2%, and the top 10% above 4.7% (Littledata, 2023). A luxury store at 1.3% can be outperforming a food store at 4%.

Ask “what is a good conversion rate” and most pages will hand you a single number. That number is a fiction, and the master table at the top of this post shows why: the honest range across major panels runs 1.4% to 2.74% before you even split by industry, and industry alone spans 6.22% down to 0.94% within one panel (Shopify, category benchmarks, 2026, sample undisclosed).

So here is the honest answer. “Good” means one thing: where you sit in the distribution of stores that look like yours.

The percentile ladder

The only public dataset in our set that publishes percentiles rather than a lone average is Littledata's Shopify benchmark. It is 2023 data and it skews toward smaller stores, both caveats it discloses itself, but it gives you four rungs to stand on (Littledata, 2,800 Shopify sites, 2023):

• Median: around 1.4%. Half of the 2,800 stores sat below this.
• Top 20%: above 3.2%. More than double the median.
• Top 10%: above 4.7%. Better than triple.
• Desktop top decile: above 6.5%. The ceiling for the best stores on the friendliest device.

Littledata benchmark of 2,800 Shopify stores, 2023 data. Littledata notes its client base skews toward smaller stores. Thresholds are “above” values: top 20% means converting above 3.2%.

Two things jump out of that ladder. First, the gap between median and top 20% (1.4% to 3.2%) is enormous, which tells you the distribution has a long tail of underperforming stores dragging the average down. Second, the desktop top decile at 6.5% shows how much headroom exists when device friction is removed, which connects back to the mobile gap covered earlier: the same panel found mobile converting at 1.2% against desktop's 1.9% (Littledata, 2023).

If you sell on Shopify and convert above 3.2% blended, you were beating four out of five comparable stores in that panel. That is a defensible definition of good. “Above 2% because a blog said so” is not.

Good for beauty is terrible for food, and heroic for luxury

Percentiles only work inside a vertical. Recall the industry rows from the table above: food and beverage stores averaged 6.22% while luxury and jewelry averaged 0.94% (Shopify, 2026), and the second panel showed the same shape, with beauty at 5.37% and luxury at 0.71% (Dynamic Yield, rolling 12-month figure, retrieved July 2026).

Run the comparison. A luxury jewelry store converting at 1.3% sits roughly 40% above its Shopify-panel industry average of 0.94%. A food store converting at 4% sits about a third below its industry average of 6.22%. The luxury store's number looks worse in absolute terms and is dramatically better in relative terms. Its buyers research for weeks, compare across sites, and spend hundreds per order. The food buyer reorders coffee in ninety seconds. Same metric, different games.

This is why cross-industry comparison produces nothing but bad decisions. If your dashboard says 1.8% and a competitor brags about 4%, the first question is what they sell, the second is their device mix, and only then does the number mean anything.

Beat your own baseline, not the average

Here is the house position: the only benchmark that pays your bills is your own trailing baseline. Percentile ladders tell you whether to be ambitious. Your baseline tells you whether you are improving.

The arithmetic makes the case better than any argument. Take a store with 8,000 sessions a month and a $72 average order value, converting at the Shopify-panel median of 1.4% (Littledata, 2023). That is 112 orders and $8,064 a month. Lift conversion by 0.2 points to 1.6%, a gain small enough that most dashboards would shrug at it, and you get 128 orders and $9,216 a month. That 0.2-point move is worth $1,152 a month, $13,824 a year, at those inputs. No benchmark table required, just your own traffic, your own AOV, and a change you can verify.

Framed as revenue per visitor, the same move takes you from $1.01 to $1.15 per session. RPV is the version of this metric worth watching, because it captures order value changes that raw conversion rate hides, and we cover why revenue per visitor beats conversion rate as a decision metric in the A/B testing guide. If you want to instrument it properly, start with how to measure revenue per visitor.

The percentile ladder answers “should I care?” Your baseline answers “did it work?” Keep the two jobs separate and both numbers become useful.

Why the benchmarks disagree: a methodology autopsy

TL;DR The published averages run from 1.4% to 2.74% because the panels behind them measure different stores, different denominators, different years, and different windows. None of them is wrong. A benchmark without a sample, a date, and a denominator is a rumor with a decimal point.

Every number in this post comes from a panel, and every panel has a shape. This section takes the four benchmark sources from the master table apart so you can see exactly why they disagree, because the disagreement is mechanical, and once you see the mechanics you can read any benchmark page, including this one, with the right suspicion.

The five mechanical reasons published averages differ

1. Panel skew. Littledata's 1.4% comes from its own client base of 2,800 Shopify sites, and Littledata itself notes that base skews toward smaller stores (Littledata, 2023). Dynamic Yield's 2.74% comes from Dynamic Yield customers, an enterprise personalization vendor owned by Mastercard, so its panel skews large (retrieved July 2026). Bigger stores have bigger optimization teams, more brand searches, and more returning buyers. The 1.34-point gap between the two figures is mostly a description of who is in each panel.

2. Denominator definition. A “conversion rate” needs a bottom half. Visits, sessions, and unique visitors produce different numbers from identical stores. Statista's figure counts 1.6% of global ecommerce visits converting in Q3 2025 (Statista, as cited by Shopify); other panels do not always say what they divide by. When a benchmark page omits its denominator, you cannot compare it to your own analytics, full stop.

3. Vintage. The Littledata data is from 2023. The Statista figure is Q3 2025. Shopify's category numbers cover its “past 12 months” as of a 2026 article. Comparing your July 2026 dashboard to a 2023 panel is comparing across three years of market change, and the honest move is to say so in the row, which is why every row in the master table carries its year.

4. Rolling windows. Dynamic Yield's benchmark is a rolling 12-month average that changes monthly, which is why this post stamps it “retrieved July 2026” rather than pretending it is a fixed fact. Baymard's 70.22% cart abandonment figure is an average of 50 studies that Baymard updates as new studies land (Baymard Institute, retrieved July 2026). Quote either without a date and you are quoting a number that may no longer exist on the source page.

5. Survivorship. Benchmark panels are built from stores that are still operating and still customers of the panel owner. Stores that failed, churned off the platform, or never got traction fall out of the data. Every panel average is therefore an average of survivors, which pulls it above whatever the true all-stores figure would be. No public panel we fetched discloses an adjustment for this.

The autopsy, panel by panel

Littledata (1.4% Shopify average). Strengths: disclosed sample size (2,800 sites), disclosed platform, percentiles published, device split published. Weaknesses: 2023 vintage and a self-acknowledged small-store skew. This is the most transparently documented panel in the table, which is exactly why its number is the lowest. Transparency and flattery rarely travel together.

Dynamic Yield (2.74% global). Strengths: refreshed monthly, split by region and vertical. Weaknesses: no disclosed sample size anywhere on the page, an enterprise-skewed customer base, and a rolling window that quietly rewrites history each month (retrieved July 2026).

Shopify (category benchmarks, 6.22% food down to 0.94% luxury). Strengths: platform-native data at category granularity. Weaknesses: no disclosed sample size or methodology on the page, and, to Shopify's credit, the article itself calls the idea of a universal ecommerce benchmark a fallacy. We quote its numbers with that caution attached.

Statista (1.6% of global visits, Q3 2025). The primary source is paywalled, so we saw this figure on Shopify's page rather than on Statista's, and we cite it exactly that way: Statista, as cited by Shopify. A citation chain you cannot walk to the end is a citation you disclose.

Holding ourselves to the same standard

It would be cheap to autopsy everyone else's methodology and hide our own, so here are our numbers. This site is about 5 weeks old in Google's index, with 2,611 impressions and 16 clicks all-time (Google Search Console via Ahrefs, pulled July 14, 2026). Sixteen clicks. We publish that because it is true, because small numbers with a source beat big numbers without one, and because a site that sells statistical honesty forfeits the right to round up. You can read more about who runs this site on the about page. For the Shopify-specific statistical picture, our sourced Shopify CRO statistics roundup applies this same standard to a different question.

How to read any benchmark: the checklist

Before a benchmark number enters your planning document, run it through six questions:

• Named source? If the page says “industry data suggests,” close the tab.
• Disclosed sample size? “2,800 stores” is checkable. “Our data” is not.
• Vintage stated? A benchmark without a date is a rumor.
• Denominator defined? Visits, sessions, or visitors, and does it match your analytics?
• Rolling or fixed window? Rolling figures need a retrieval date attached.
• Survivorship acknowledged? Almost never, but the panels that mention their own skew (Littledata does) earn extra trust.

Where this post will rot

Five rows in the master table are rolling: Dynamic Yield's two benchmark pages (conversion and order value), Baymard's abandonment meta-average, GoodUI's test counts, and Statcounter's device share. All were re-fetched on July 14, 2026 and all will drift. Our policy is to re-pull every rolling source on each annual refresh of this page and restamp the retrieval dates, and the Updated label at the top tells you when that last happened. If you catch a stale figure between refreshes, the source link next to it takes you to the live number.

How to calculate your ecommerce conversion rate (and how much noise is in it)

TL;DR Sessions with a completed order, divided by total sessions, times 100. Then three things corrupt the reading: the denominator you chose, sampling noise (a true 2% rate on 1,000 sessions can legitimately read anywhere from 1.1% to 2.9%), and shifts in your traffic mix.

The autopsy above covered everyone else's numbers. This section is about yours, because before your rate can be compared to anything, it has to be computed honestly, and there are three ways the computation quietly lies to the person running it.

The formula, and what Shopify actually divides by

The standard calculation: conversion rate = sessions that included a completed order ÷ total sessions × 100. Shopify's own analytics works exactly this way. Its conversion rate breakdown report computes each funnel step as “the number of sessions for a step divided by the number of total sessions,” and defines “sessions that completed checkout” as sessions where a customer purchased (Shopify Help Center, behavior reports, retrieved July 2026).

Two details in that definition earn their keep. First, the unit is the session, not the order. Shopify's documentation notes that a customer can place more than one order inside a single session, so your order count and your converting-session count are different numbers, and a rate computed as orders divided by sessions will read slightly higher than the one your platform reports. Second, sessions are not the only possible denominator. Statista's 1.6% counts visits (Statista, as cited by Shopify), and some panels report per-visitor rates, where one shopper's three visits collapse into a single denominator unit. The difference is not cosmetic. A store with 100 converting sessions out of 6,000 sessions reads 1.67%. If those 6,000 sessions came from 4,400 unique visitors, the same month measured per visitor reads 2.27%. Same store, same behavior, 0.6 points apart, and neither number is wrong. They answer different questions.

The rule that follows: pick one denominator, write it down, and never compare your sessions-based rate to a visitor-based benchmark without saying so. Two smaller hygiene rules ride along. Exclude your own staff and known bots the same way every month, because a filter change masquerades as a conversion change. And if you run two analytics tools on the same store, expect their session counts to disagree; different session timeouts and bot filters guarantee it, so pick one tool as the system of record and benchmark only against itself.

The noise floor: how much your rate wobbles when nothing changes

A conversion rate is a proportion measured on a sample, which means it carries a margin of error like any poll. The arithmetic is standard binomial math, shown in the table note, and the result surprises most merchants who run it for the first time.

Sessions in the window	95% range around a true 2%	What the dashboard can show you
1,000	1.13% – 2.87%	“Conversion collapsed” or “conversion soared,” from nothing
5,000	1.61% – 2.39%	±0.4 points of pure noise
25,000	1.83% – 2.17%	±0.17 points
100,000	1.91% – 2.09%	The second decimal starts meaning something

Derived, not fetched: 95% interval = 2% ± 1.96 × √(0.02 × 0.98 ÷ n), the standard error of a proportion. The exact band moves with your true rate; the shape of the table does not.

Read the first row again. A store with 1,000 sessions a month and a genuinely constant 2% conversion rate will, purely by chance, post some months near 1.2% and others near 2.8%. No redesign happened. No checkout broke. The dice landed differently. If your store runs low thousands of sessions, month-over-month conversion moves smaller than about half a point are unreadable, and reacting to them is steering by static.

The same arithmetic sets the price of A/B testing, which gets itemized in full a few sections down. For benchmarking, the practical rule is simpler: pull at least a full month, and if your traffic is small, pull a quarter and accept that your “true” rate is a band, not a point.

Windows and seasonality: compare July to July, never July to November

A conversion rate is stamped with the period it was measured in, and periods are not interchangeable. Holiday traffic converts differently from January traffic, a sale week converts differently from the week after it, and a rate that blends Black Friday into a quarterly average tells you about your calendar, not your site. The clean comparison for your own store is the same window a year earlier, not the adjacent month.

Even year-over-year needs one adjustment: the whole field drifts. Contentsquare's benchmark, one of the largest published panels, recorded conversion rates falling 6.1% year over year in its 2025 report (90 billion sessions) and another 5.1% in its 2026 edition (6,500+ sites, 99 billion sessions, retrieved July 2026). Both figures come from Contentsquare's enterprise-skewed customer base, with all the panel caveats from the methodology section, but two consecutive market-wide declines mean a store converting exactly as well as it did last year has, relative to the field, gained ground. If your July 2026 rate sits 3% below your July 2025 rate, the honest first reading is “roughly market,” not “something broke.”

The mix trap: your rate can fall while every segment improves

The composition effect from the platform section applies to your own dashboard, and it is worth watching happen in digits. Recall Littledata's device rates: desktop 1.9%, mobile 1.2% (2,800 Shopify stores, 2023). Take a store at exactly those rates with a 50/50 device split. Blended conversion: 0.5 × 1.9 + 0.5 × 1.2 = 1.55%.

Now the store runs a successful mobile ad campaign and the split moves to 30/70 desktop-to-mobile. Same site, same segment rates: 0.3 × 1.9 + 0.7 × 1.2 = 1.41%. The dashboard reports conversion down 9%. Nothing about the store changed except who showed up.

It gets stranger. Suppose the store also improves both segments, desktop to 2.0% and mobile to 1.3%, while the mix shifts further to 25/75: 0.25 × 2.0 + 0.75 × 1.3 = 1.48%. Every segment now converts better than at the start, and the blended rate still fell, 1.55% to 1.48%. This is Simpson's paradox wearing an ecommerce apron, and it is not an edge case. It is what happens, mechanically, every time a store buys cheaper mobile or international traffic while improving its site.

The defense costs one habit: keep a small rate card, conversion by device and by traffic source, and when the blended number moves, check the mix before you check the site. A falling blended rate over stable segment rates is a marketing-mix event, not a CRO emergency. With your own number computed honestly, the next question is what actually moves it.

What actually moves conversion rates: the evidence, with caveats attached

TL;DR Site speed and checkout design are the two best-documented levers in public evidence. A site loading in 1 second converted at 2.5x the rate of one loading in 5 seconds (Portent, 2022), and Baymard models a 35.26% conversion gain from better checkout design (modeled, retrieved July 2026). Every number below carries an evidence grade, because most blogs quote the modeled ones as if they were measured.

Knowing where you sit in the distribution is diagnosis. This section is the treatment evidence, and it comes with a grading system most benchmark posts skip: experimental (a controlled test measured it), correlational (fast sites also convert better, but fast sites differ in other ways too), and modeled (someone did arithmetic on top of research, and the output is a projection). The grade matters more than the headline number.

Lever 1: site speed (correlational, two independent sources)

The most-cited public speed data comes from Portent's study of 20 websites, of which only 6 were ecommerce, covering just over 100 million page views (originally run in 2019, updated with fresh data in 2022). Ecommerce sites loading in 1 second converted at 3.05%, at 2 seconds 1.68%, and at 4 seconds 0.67%, which is where the headline finding comes from: a 1-second site converts at 2.5x the rate of a 5-second site (Portent, 2022). The caveats are real: 6 ecommerce sites is a small panel, the data is 2022, and the finding is correlational. Fast sites tend to be run by teams that are careful about everything else too.

The second source is stronger on method and weaker on disclosure. Deloitte's “Milliseconds Make Millions” analyzed four weeks of mobile data from retail, travel, and luxury brands across Europe and the US and found that a 0.1-second improvement in mobile site speed was associated with retail conversions rising 8.4% and retail average order value rising 9.2% (Deloitte, 2020). Two caveats travel with it: the study was commissioned by Google, which has an interest in a fast web, and the summary page does not disclose site or session counts.

Two independent panels, two vintages, one direction. Speed is the lever with the most consistent public evidence behind it, and it is also the one most stores can act on this week: compress images, cut app scripts, measure again.

Lever 2: checkout design (measured problem, modeled fix)

The problem is measured. Across 50 separate studies of cart abandonment, the average documented rate is 70.22% (Baymard Institute, meta-average of 50 studies, retrieved July 2026). Seven of ten carts, abandoned. That number is an average across studies with varying methodologies, and Baymard updates it as new studies arrive, hence the retrieval date.

The fix is modeled, and this distinction matters. Baymard states that the average large ecommerce site can gain a 35.26% increase in conversion rate through better checkout design, and that roughly $260 billion in lost orders is recoverable across the US and EU (Baymard, retrieved July 2026). Those figures are projections from Baymard's checkout usability research applied to aggregate sales data. They are not the result of an A/B test anyone ran. Most articles quote 35.26% as if it were a measured outcome; it is a modeled ceiling for large sites fixing solvable checkout UX problems, and Baymard says so. We grade it modeled and quote it anyway, because even a modeled ceiling tells you the checkout is where recoverable money concentrates.

What to do with that: audit your own checkout before touching anything upstream of it, and if abandonment is your specific leak, we keep a practical walkthrough in reducing cart abandonment on Shopify. If shipping cost is the friction you suspect, the free shipping threshold calculator lets you test a threshold against your own AOV instead of guessing.

Lever 3: experience quality (correlational, one very large panel)

Contentsquare's 2025 Digital Experience Benchmark, drawn from 90 billion sessions across 6,000 sites, found that 40% of all online visits showed user frustration signals in 2024, that conversion rates dropped 6.1% year over year, and that slow-loading content drove 53% of single-page bounced sessions (Contentsquare, 2025 benchmark, Q4 2023 vs Q4 2024). The same report found that sites increasing session depth by 10% or more saw an average 5.4% conversion boost (Contentsquare, correlational).

Caveats: “frustration” is Contentsquare's proprietary signal definition, the panel is its own enterprise-skewed customer base, and the session-depth finding is a correlation. Stores whose visitors go deeper are stores visitors already like. Nobody proved that forcing depth causes conversion. Still, the 53% figure on slow content independently corroborates the speed evidence above, from a third panel with a different methodology, and three panels pointing the same direction is about as good as public correlational evidence gets.

Lever	Best public evidence	Figure	Grade
Site speed	Portent, 2022, 20 sites (6 ecommerce)	2.5x conversion at 1s vs 5s load	Correlational
Mobile speed	Deloitte, 2020, Google-commissioned	+8.4% retail conversions per 0.1s	Correlational (field analysis)
Checkout design	Baymard, retrieved Jul 2026	35.26% potential CVR gain	Modeled
Cart recovery target	Baymard, 50-study average, retrieved Jul 2026	70.22% average abandonment	Measured (meta-average)
Content speed	Contentsquare, 90B sessions, 2025	53% of single-page bounces	Correlational
Session depth	Contentsquare, same panel	+5.4% CVR with 10%+ depth gain	Correlational

Grades: measured = someone counted it; correlational = it moves together with conversion, causation unproven; modeled = a projection computed on top of research, not an observed result.

Notice what the table does not contain: a single controlled experiment proving a specific tactic lifts conversion by a specific amount for stores in general. That absence is the honest state of public evidence. The experimental record exists, but it lives inside individual companies' test programs, and what leaks out of those programs is the subject of the next section.

What won't work: copying the leader

The tempting shortcut is to skip the evidence grades, find a store converting at 4.7%, and copy its product page. The public testing record says this fails more often than it works, because effects are context-dependent: a layout that wins for a beauty brand with heavy repeat purchasing can lose for a furniture store where every buyer is new. Speed helps almost everyone. Checkout friction hurts almost everyone. Nearly everything else is a hypothesis about your store that needs a test, and the next section puts hard numbers on how often such tests actually win. Spoiler: less often than any agency deck has ever admitted.

The testing reality check: what fraction of changes actually win

TL;DR Across the best public datasets, somewhere between 8% and 33.5% of A/B tests produce a significant winner, and the median lift across all tests is a rounding error. Plan for one winner per three to ten tests.

Every benchmark section so far has an implied next step: your store sits below some row in the table, so you change things until it doesn't. Here is the part most benchmark posts skip. Most changes do nothing measurable, and the public evidence on this is unusually good.

Start with the practitioners who run the most experiments on Earth. Ronny Kohavi, who built and ran experimentation at Microsoft and Airbnb, puts the median organization at roughly 10% of experiments moving the metric they were designed to improve, with Microsoft at 33%, Bing at 15%, and Airbnb Search at 8% (Kohavi, interviewed on the GrowthBook blog). Note the direction of that spread. Microsoft's 33% is not because Microsoft is worse at product than Airbnb. It reflects heavy pre-experiment screening, where weak ideas die before they ever reach a test. Bing and Airbnb Search are mature, heavily optimized surfaces where the easy wins were taken years ago. A young store that has never tested anything sits closer to the cold-start end, where win rates run higher because the obvious problems are still unfixed.

Now the independent datasets. Georgi Georgiev analyzed 1,001 real A/B tests run through his Analytics-Toolkit platform and found 33.5% ended as statistically significant winners (Analytics-Toolkit, “What Can Be Learned From 1,001 A/B Tests?”, 2022). Before you anchor on that number, read the caveat the way we would want you to read ours: Georgiev describes his users as advanced and expert CRO practitioners. That panel is the top of the skill curve, so 33.5% is a ceiling, not an expectation. The same dataset holds the two numbers that should actually change your planning. The median lift across all 1,001 tests was 0.08%. The mean was 2.08%. Among the winners, the median lift was 7.5% and the mean 15.9%, and both winner figures should be read with the winner's curse in mind, the tendency of statistically significant results to overstate the true effect. Average test duration: 35.4 days (the median was 30). And 88% were simple A/B tests, one variant against control.

GoodUI's evidence base tells a similar story from a different panel. As re-fetched on July 14, 2026, it aggregates 633 tests across 147,071,944 visitors, contributed by real teams: 165 winning, 43 losing, and 425 statistically insignificant (GoodUI Evidence; counts grow over time, figures as retrieved). That is roughly 26% significant winners, and it means the modal outcome of an A/B test, in a self-selected database where people volunteer their results, is “nothing happened.”

Win rate = share of tests with a statistically significant winner. Panels differ in maturity and skill: expert-run panels and heavily screened programs sit high, mature optimized surfaces sit low. See caveats in prose.

So why test at all, if two thirds to nine tenths of tests come back flat? Because the distribution has a fat right tail, and you cannot know in advance which idea is in it. The canonical example: a Bing engineer's ad-headline change, one of hundreds of proposed ideas, sat in a backlog for over six months because the program managers deemed it low priority. When it finally ran, it increased Bing's revenue by 12%, more than $100 million per year in the US alone (Kohavi & Thomke, Harvard Business Review, September 2017). The experts had already ranked that idea as low priority. That is the whole argument in one anecdote: expert judgment about which changes will win is bad, effects are discovered rather than predicted, and the only way to find the +12% idea is to run it alongside the duds.

Put the numbers together and a realistic expectation forms:

• If you are new to testing with obvious friction on the site, expect the higher end, something like one winner in three or four tests.
• If your store is already tuned, expect the Bing/Airbnb end, one in seven to twelve.
• Whatever your rate, most winning effects will be single digits. The median winner in the 1,001-test dataset lifted its metric 7.5%, not 50%.

This is also why one redesign almost never closes a benchmark gap. If your store converts at 1.4% and the beauty row in the industry table says 4.94%, the distance between those numbers is 3.5 points, roughly a 250% relative gain. The median measured effect across a thousand real tests was 0.08%. Gaps that size close through industry mix, traffic mix, and years of compounding small wins, not through a hero project.

Two practical consequences. First, at an average duration of 35.4 days per test (Analytics-Toolkit, 2022), a store running sequential tests gets maybe ten shots a year, so screening ideas against evidence, like the graded levers in the previous section, is worth real money. Second, low win rates are the argument for honest statistics, not against testing. If only a quarter of tests genuinely win but your tool declares winners half the time, half your “wins” are noise you are now shipping to production. We wrote up how to hold that line in the statistical significance section of our A/B testing guide, what to do when your traffic makes 35-day tests painful in the low-traffic playbook, and why we think honest A/B testing is a product feature rather than a footnote. If you are on Shopify and wondering whether the platform's built-in testing covers this, here is what Shopify Rollouts can and can't do.

What moving your conversion rate actually costs: the power problem

TL;DR Roughly 70% of 115 public A/B tests were too small to reliably detect anything under a ~21% lift (Georgiev, 2018), while the median real winner lifts 7.5% (Analytics-Toolkit, 2022). Detecting a 10% lift on a 2% baseline honestly costs about 157,000 sessions.

The win rates above say most tests lose. There is a harder truth underneath them: most tests, as actually run, could not have detected a realistic win in the first place.

The evidence comes from the same researcher behind the 1,001-test dataset. In 2018, Georgi Georgiev pulled 115 publicly available A/B tests from GoodUI's free evidence database and recomputed their statistics. Roughly 70% of them, 80 of 115, ran with low statistical power, and the median test could only reliably detect, at 90% power, an effect of 20.84% or larger. The average threshold was worse, 27.8% (Georgiev, Analytics-Toolkit, “Analysis of 115 A/B Tests,” 2018; public, self-selected tests, so read the exact percentages gently).

Hold those thresholds against the outcome data. The median winning test in the 2022 dataset lifted its metric 7.5% (Analytics-Toolkit, 2022). So the typical public test was sized to detect an effect roughly three times larger than the typical winner actually delivers. A test like that is not an instrument; it is a coin flip with a dashboard attached. When it comes back “no significant difference,” it has told you close to nothing, because it could not have seen a real 7% winner had one been sitting right there.

Underpowering also corrupts the wins. A small test can only cross the significance line when its observed lift happens to be enormous, so the winners that emerge from underpowered tests are systematically the most exaggerated ones. That is the winner's curse from the section above with its mechanism exposed: run small tests and you ship fewer winners, and you overstate the ones you ship.

The sessions bill, itemized

How big is big enough? A standard shortcut, Lehr's rule, prices a two-variant test at 80% power and 95% significance: sessions per variant ≈ 16 × p(1−p) ÷ d², where p is your baseline conversion rate and d is the absolute lift you want to detect. Here is that arithmetic on a 2% baseline.

Lift you want to detect	Sessions per variant	Total sessions	At 20,000 sessions/month
+50% relative (2.0% → 3.0%)	~3,100	~6,300	Under 2 weeks
+20% relative (2.0% → 2.4%)	~19,600	~39,200	~2 months
+10% relative (2.0% → 2.2%)	~78,400	~156,800	~8 months
+5% relative (2.0% → 2.1%)	~313,600	~627,200	~2.6 years

Derived, not fetched: Lehr's rule, n ≈ 16 × p(1−p) ÷ d², at 80% power and 95% two-sided significance on a 2% baseline. It is an approximation; trust the shape, not the last digit.

This table explains several things at once. It explains the 35.4-day average duration in the expert dataset (Analytics-Toolkit, 2022): practitioners with real traffic still need five weeks, because they are hunting effects in the 10–20% band. It explains why the median detectable effect in the 2018 public-test sample landed near 21%: at typical store traffic, that is all a month-long test can honestly see. And it explains why “we tested it and nothing happened” from a 10,000-session store is usually a statement about the test, not about the change.

For a small store the playbook follows directly: run fewer, bolder tests, changes large enough to plausibly clear your own detection threshold, and ship the small tweaks with before-and-after measurement instead of pretending to test them. We walk the full sample-size arithmetic, with worked examples, in how many visitors an A/B test actually needs.

The benchmark-chasing arithmetic

Now put the whole reality check together and price the original fantasy: closing the gap to a published benchmark. Suppose a store converting at the Littledata median of 1.4% decides to reach Dynamic Yield's 2.74% average. That is a 96% relative improvement.

Give that store a genuinely good testing year: ten sequential tests at the 35-day expert cadence, a 25% win rate (mid-range of the public datasets above), and every winner landing the median 7.5% lift. Two and a half winners compounding at 7.5% each works out to roughly a 20% relative gain for the year, which takes the store from 1.40% to about 1.68%. Sustain that year after year, an assumption doing heroic work, and the 96% gap closes in about 3.7 years of uninterrupted, well-executed, honestly-measured testing. All of that is derived arithmetic on the cited inputs, and your inputs will differ, but the order of magnitude is the point.

That is the honest price tag, and it is why this post keeps insisting the benchmark is context rather than target. A store that spends those years compounding against its own baseline gets rich exactly as fast as the one chasing the panel average, with none of the despair. It is also why every one of those expensive tests should be pointed at a metric that actually pays, which is the next section's subject.

Don't celebrate add-to-carts: which metrics predict revenue

TL;DR Add-to-cart lift explains only about a quarter of the variance in actual order lift (R = 0.4983 across 44 experiments, GoodUI). The closer a metric sits to the purchase, the better it predicts revenue.

There is a quieter failure mode than losing a test: winning one on the wrong metric.

Plenty of “conversion” wins are measured on add-to-cart rate, because ATC events are frequent and tests on them reach significance fast. The problem is that ATC lift and order lift are loosely coupled. GoodUI ran the correlation across its database and found add-to-cart lift correlates with order lift at R = 0.4983 across 44 experiments, and at R = 0.4349 across roughly 200 experiments from Conversion.com's dataset (GoodUI, “Do adds to cart or progression metrics correlate with sales in A/B tests?”). Square those correlations and the picture sharpens: an R of 0.4983 means ATC movement explains about 25% of the variance in sales movement (R² = 0.248). The other three quarters is everything ATC does not see, including shoppers who add to cart because your new button is shinier and then bounce at shipping costs.

The same study checked metrics further down the funnel. Generic progression metrics (moving to any next funnel step) correlate with sales at R = 0.5085 across 119 experiments. Checkout visits correlate with orders at R = 0.6085 across 533 experiments, with a p-value near zero, on Conversion.com data (same GoodUI post). Lay the four datasets side by side and the pattern is monotonic:

Metric measured	Correlation with order lift (R)	n (experiments)	R² (share of sales variance explained)
Add to cart (GoodUI)	0.4983	44	~25%
Add to cart (Conversion.com)	0.4349	~200	~19%
Funnel progression	0.5085	119	~26%
Checkout visits	0.6085	533	~37%

Source: GoodUI correlation study, observational meta-analysis across contributed A/B tests, retrieved July 2026. R² computed from the published R values.

Both analyses are observational, correlations across contributed tests rather than a controlled study, so treat the exact decimals gently. The direction, though, is consistent across four samples: the closer a metric sits to money changing hands, the better it predicts money changing hands. Even checkout visits, the best proximate metric in the table, explains barely over a third of sales variance.

For a merchant, this cashes out into two rules. First, when a case study reports a “conversion lift,” ask which conversion. An ATC win with unknown revenue impact is, per the numbers above, closer to a coin flip on sales than to a sure thing. Second, benchmark and test on revenue-proximate metrics, ideally revenue per visitor, because a variant can raise conversion rate while quietly shrinking order values, a trap we dissected in the sample-size section of our product image testing post.

This table is also the statistical spine of a position we hold about AI in testing: language models are good at proposing hypotheses and terrible at being trusted with verdicts, and the verdict should come from deterministic math on revenue-proximate metrics. We wrote the full argument in who decides in AI A/B testing, and the mechanics of how a winner actually gets shipped follow from it.

Conversion rate vs revenue per visitor: when a lower rate is the right answer

TL;DR Revenue per visitor = conversion rate × average order value. Dynamic Yield's panel prices the average session at $5.07 (2.74% × $185, both retrieved July 2026). Luxury's $386 AOV only half-rescues its 0.71% conversion rate, and price or shipping tests can win on revenue while losing on conversion.

Conversion rate has no dollar sign on it. Revenue per visitor does, and it is the metric this whole post has been quietly steering toward.

The formula is one multiplication: RPV = conversion rate × average order value. Dynamic Yield publishes both halves for the same panel, which makes the arithmetic checkable: a 2.74% conversion rate times a $185 global average order value (Dynamic Yield conversion and AOV benchmark pages, both rolling 12-month, retrieved July 2026) prices the average session in that panel at $5.07. Derived, and worth deriving for your own store before reading on: your rate, times your AOV, per session.

Cut	Segment	Average order value
Global	Whole panel	$185
Region	EMEA	$208
Region	Americas	$157
Region	APAC	$119
Device	Desktop	$255
Device	Mobile	$164
Device	Tablet	$161
Industry	Luxury & jewelry (panel high)	$386
Industry	Pet care & veterinary (panel low)	$63

Dynamic Yield AOV benchmark page, 12-month rolling averages in USD per order, retrieved July 2026. Same panel and caveats as its conversion page: sample undisclosed, enterprise-skewed, changes monthly.

The luxury test: does AOV really rescue a low conversion rate?

The industry section made the standard argument: luxury's sub-1% conversion is fine because order values carry the revenue. Dynamic Yield's data lets us check how far that logic actually stretches inside one panel. Luxury and jewelry converts at 0.71% and its orders average $386, the highest AOV in the panel (both Dynamic Yield, retrieved July 2026). Multiply: $2.74 per session, against the panel-wide $5.07.

So the comforting story is only half true. Luxury's AOV premium runs about 2.1x the global average, but its conversion deficit runs 3.9x, and 2.1 does not cancel 3.9. In this panel, the average luxury session earns roughly 54% of what the average session earns. That does not make luxury stores broken. It means their economics run on levers this metric barely sees: repeat purchase over a multi-year customer relationship, email capture across a weeks-long consideration window, and order value itself. But it retires the lazy version of the claim, that a high ticket automatically pays for a low rate. Sometimes it does not even come close, and only the multiplication tells you. (All derived from the two Dynamic Yield pages; panel arithmetic, not a law of retail.)

The device version: most of the traffic, less than half the money

Statcounter's live tracker puts mobile at 51.51% of worldwide web traffic against 47.12% for desktop (Statcounter Global Stats, June 2026; all web traffic, not ecommerce-specific). Devices differ on both halves of the RPV multiplication: Littledata's panel converts desktop at 1.9% versus mobile's 1.2% (2023), and Dynamic Yield's panel prices desktop orders at $255 versus mobile's $164 (retrieved July 2026).

Cross the two and you get an illustration, flagged as one because it multiplies figures from two different panels: desktop, 1.9% × $255 ≈ $4.85 per session; mobile, 1.2% × $164 ≈ $1.97. The conversion gap between devices is 1.6x. The revenue-per-session gap is 2.5x, because mobile loses on both terms of the multiplication at once. If mobile is the majority of your traffic, which the Statcounter share suggests it is for most stores, then mobile work is worth considerably more than its conversion gap alone implies, and the speed evidence in the levers section picks up a second engine.

When the right move lowers your conversion rate

Here is the counterintuitive payoff of the RPV lens. Take a store at 2.0% conversion and a $60 AOV: $1.20 per session. It raises its free-shipping threshold and nudges prices, conversion slips to 1.8%, and AOV climbs to $70: $1.26 per session. Judged on conversion rate, the change failed, down 10%. Judged on revenue per visitor, it won, up 5%, on 10% fewer orders to pick, pack, and ship. Hypothetical arithmetic with stated inputs, and the margin side gets friendlier still; we work that half in the contribution margin guide.

The pattern shows up at market scale too. Contentsquare's 2026 benchmark recorded conversion rates down 5.1% year over year while average order values rose 6% (6,500+ sites, 99 billion sessions, retrieved July 2026). Multiply the two aggregate moves and revenue per visit lands within a rounding error of flat, up about 0.6%, in a year that every conversion-only dashboard scored as a defeat. Multiplying two panel-level medians is an illustration rather than any single store's ledger, but the direction of the lesson holds: a conversion rate can be bought with discounts and sold with price discipline. Revenue per visitor has to be earned.

Two honest limits before this turns into a slogan. RPV is a noisier metric than conversion, because order values vary far more than order counts, so a test judged on RPV needs more traffic to reach a verdict, the same power arithmetic as above with a bigger bill. And conversion rate remains the right diagnostic inside the funnel, where the question is where sessions leak rather than what a session is worth. Use conversion to find the leak, use RPV to decide whether the fix was worth shipping. Price changes are where the difference bites hardest, and the Shopify price testing guide is the practical companion.

How to benchmark your own store honestly: the worked method

TL;DR Six steps: fix your denominator, segment by device and industry, compare percentile to percentile at matching vintage, pick an evidence-graded lever, validate with realistic win-rate expectations, and track revenue per visitor.

Everything above compresses into a method you can run this afternoon. No step introduces a number that has not already appeared in this post with its source attached.

1. Fix your denominator before you look at anything. Decide whether you are measuring orders per session or orders per visitor, and write it down. The published benchmarks disagree with each other partly because they do not share a denominator, and your comparison inherits that problem the moment you get sloppy about yours. Pull at least a full month so weekday and weekend mix is not distorting the figure. The full arithmetic, including how much sampling noise a small store's monthly rate carries, is in the calculation section above.
2. Segment by device, then find your industry row. A blended rate hides the split that matters most. Shopify stores in Littledata's panel converted at 1.2% on mobile versus 1.9% on desktop (Littledata, 2,800 Shopify stores, 2023), so compute yours separately, and note your traffic mix. Then pick your row from the industry table: a store selling supplements should be reading the food and beverage row, not the global average.
3. Compare percentile to percentile, at matching vintage. Put your number against the ladder: Shopify-panel median 1.4%, top 20% above 3.2%, top 10% above 4.7%, desktop top decile above 6.5% (Littledata, 2023). Remember those figures are 2023 vintage from a small-store-skewed panel. If you are at 2.1% in home and furniture, whose Shopify-published category benchmark is 1.41% (Shopify blog, 2026), you are ahead of your row even though you are “below average” globally. The comparison that matters is inside your vertical and device mix.
4. Pick the highest-grade lever you have not pulled. From the evidence table earlier in this post: speed and checkout UX carry the strongest documentation, correlational but replicated. Frustration signals and modeled checkout gains rank below them. Do not start with the lever that has the biggest headline number; start with the one whose evidence grade is highest and whose problem you can confirm exists on your own site.
5. Validate the change with realistic expectations. Tests in the 1,001-test dataset averaged 35.4 days (Analytics-Toolkit, 2022), and win rates across the public datasets run from 8% to 33.5% depending on panel and maturity. If your first two tests come back flat, that is the distribution behaving normally, not proof that testing is useless. Budget for a program, not a miracle.
6. Track revenue per visitor, not just conversion rate. Per the correlation table above, even checkout visits only explain about 37% of sales variance (GoodUI, 533 experiments). A conversion-rate win that shrinks average order value can be a revenue loss. RPV catches that; conversion rate alone does not.

Then repeat, quarterly. Re-pull your baseline, re-check the rolling benchmarks (they move; that is why every rolling figure in this post carries a retrieval date), and pull the next lever. The stores in the top decile of that percentile ladder did not get there in one redesign. At a median winner lift of 7.5% (Analytics-Toolkit, 2022), the ladder is climbed in compounding single-digit steps.

A disclosure, since this post has been holding everyone else to a sourcing standard and should hold itself to one too. We are building StorePilot, an AI CRO tool for Shopify that automates most of this method: it watches visitor behavior, proposes evidence-ranked tests, and calls winners with the deterministic statistics this post keeps insisting on. It has not launched yet; there is a waitlist, and what it will cost is public. That is the extent of the pitch. The math in these six steps works the same whether or not you ever use our tool.

Questions merchants keep asking

What is the average ecommerce conversion rate?

There is no single figure. Dynamic Yield's rolling 12-month benchmark shows a 2.74% global average (retrieved July 2026), while Statista reports 1.6% of global ecommerce visits converted in Q3 2025 (as cited by Shopify). The spread comes from panel composition and methodology, not measurement error.

What is a good conversion rate for ecommerce?

“Good” is a percentile inside your vertical, not a universal number. In Littledata's benchmark of 2,800 Shopify stores (2023), the top 20% converted above 3.2% and the top 10% above 4.7%. A luxury store at 1.3% can outperform a food store at 4% relative to its industry.

What is the average Shopify store conversion rate?

1.4%, per Littledata's benchmark of 2,800 Shopify stores (2023 data). Note the panel caveat: Littledata's client base skews toward smaller stores, which pulls this below all-ecommerce averages like Dynamic Yield's 2.74%. The gap is a composition effect, not a Shopify penalty.

Why is mobile conversion lower than desktop?

In Littledata's Shopify panel (2023), mobile converted at 1.2% versus 1.9% on desktop. Speed is one documented driver: Deloitte's “Milliseconds Make Millions” study (2020, Google-commissioned) found a 0.1-second mobile speed improvement lifted retail conversions 8.4%. Smaller screens and interrupted sessions do the rest.

What percentage of A/B tests win?

Between roughly 10% and 33.5%, depending on who is running them. Kohavi puts the median organization near 10% (Microsoft 33%, Bing 15%, Airbnb Search 8%). Analytics-Toolkit's analysis of 1,001 expert-run tests found 33.5% significant winners. GoodUI's database sits near 26%.

What is the average cart abandonment rate?

70.22%, per Baymard Institute's running average of 50 different cart-abandonment studies (retrieved July 2026). The figure is a meta-average across studies with varying methodologies, and Baymard updates it over time, so quote it with its retrieval date.

Is a 2% conversion rate good?

It depends on your industry row. 2% sits above the 1.4% Shopify-panel median (Littledata, 2023) but below the 2.74% all-ecommerce average (Dynamic Yield, retrieved July 2026). For luxury and jewelry, where Shopify's benchmark is 0.94%, 2% would be exceptional.

How do I increase my ecommerce conversion rate?

Start with the two best-documented levers: site speed (Portent's 2022 data found 1-second pages converting at 2.5x the rate of 5-second pages) and checkout UX (Baymard models a 35.26% potential gain, a modeled figure, not a measured one). Then A/B test rather than assume; see the evidence section above.

How do you calculate ecommerce conversion rate?

Divide sessions that included a completed order by total sessions, then multiply by 100. Shopify's analytics uses exactly this session-based definition (Shopify Help Center, retrieved July 2026). Use at least a full month of data, keep the same denominator over time, and note that a 2% reading on 1,000 sessions carries roughly a ±0.9-point margin of error.

Sources

Every number in this post traces to one of the sources below, with its retrieval date. Rolling benchmarks change over time; figures are as retrieved.

1. Ronny Kohavi, interviewed on the GrowthBook blog – org-level experiment win rates. Retrieved July 14, 2026.
2. Kohavi & Thomke, “The Surprising Power of Online Experiments”, Harvard Business Review, September 2017 – the Bing +12% headline test. Retrieved July 14, 2026.
3. Georgi Georgiev, “What Can Be Learned From 1,001 A/B Tests?”, Analytics-Toolkit, 2022 – win rates, lift distributions, durations. Retrieved July 14, 2026.
4. Georgi Georgiev, “Analysis of 115 A/B Tests: Average Lift is 4%, Most Lack Statistical Power”, Analytics-Toolkit, 2018 – underpowering rates and minimum detectable effects. Retrieved July 14, 2026.
5. GoodUI Evidence page – 633-test outcome counts. Counts grow monthly; figures as retrieved July 14, 2026.
6. GoodUI correlation study – metric-to-sales correlations. Retrieved July 14, 2026.
7. Baymard Institute, cart abandonment statistics – 70.22% meta-average; 35.26% and $260B modeled checkout figures. Rolling; retrieved July 14, 2026.
8. Littledata, average ecommerce conversion rate – 2,800 Shopify stores, 2023 data. Retrieved July 14, 2026.
9. Shopify blog, ecommerce conversion rate – industry category benchmarks, 2026; sample undisclosed. Retrieved July 14, 2026.
10. Statista, Q3 2025 global conversion figure – as cited by Shopify (primary is paywalled). Retrieved July 14, 2026.
11. Dynamic Yield (Mastercard) conversion benchmark – rolling 12-month; changes monthly; figures as retrieved July 14, 2026.
12. Dynamic Yield (Mastercard) average order value benchmark – rolling 12-month AOV by region, device, and industry; changes monthly; figures as retrieved July 14, 2026.
13. Portent site speed study, updated 2022 – 20 sites, 6 ecommerce. Retrieved July 14, 2026.
14. Deloitte, “Milliseconds Make Millions”, 2020 – Google-commissioned. Retrieved July 14, 2026.
15. Contentsquare, 2025 Digital Experience Benchmark press release – 90B sessions. Retrieved July 14, 2026.
16. Contentsquare, 2026 Digital Experience Benchmark – 99B sessions, 6,500+ sites; headline figures from the ungated summary page (full report gated). Retrieved July 14, 2026.
17. Statcounter Global Stats, platform market share – mobile vs desktop share of worldwide web traffic, June 2026; changes monthly. Retrieved July 14, 2026.
18. Shopify Help Center, behavior reports – the session-based conversion rate definition. Retrieved July 14, 2026.
19. Our data: Google Search Console via Ahrefs for usestorepilot.com, pulled July 14, 2026.