Skip to content
Original research

Shopify return fraud statistics 2026: what 73,000+ orders revealed

Last reviewed ·By Adrien Bokor, founder of RefundSentry

Most return fraud statistics recycle the same industry surveys. This page is different: we analyzed 73,000+ orders, 2,100+ refunds and about 860,000 EUR of refunded value from one anonymized Shopify store over 12 months in 2025–2026, including roughly 70 bank disputes. Public counts and amounts are rounded to reduce re-identification risk; percentages and ratios use the full-precision source data, so displayed figures may not recompute exactly. No row-level data is published, and the null results appear alongside the findings. In September 2026 a second store (48,000 orders, USD) was run through the same method to test which findings hold across catalogs.

Start Free 14-Day TrialJump to the findings

Want this analysis on your own store? Start a 14-day free trial through Shopify.

The findings

Nine findings, each with the number, the context, and what it means for a merchant reading their own refund line. Unless stated otherwise, percentages use shipped-order refunds as the denominator (see methodology).

46% of refunded value was on orders that never shipped

Almost half the refund line never involved a shipped parcel

Of about 860,000 EUR refunded over the year, roughly 400,000 EUR (about 800 refunds) sat on orders that were never fulfilled. Most were formally cancelled before shipment. That money is pre-shipment cancellations: payment issues, address problems, or buyer remorse caught early. It is not returns abuse.

The post-shipment refund surface is the other half: about 460,000 EUR across roughly 1,300 refunds on orders that actually shipped. Every post-shipment statistic below uses that slice as its denominator. A refund dashboard that mixes the two halves into one number overstates the amount exposed after fulfillment by roughly 2x and hides where the patterns live.

#never-shipped

100% of observed payment disputes came from orders Shopify's fraud analysis rated LOW risk

Shopify's fraud analysis rated every order that became a chargeback as low risk

Shopify's native fraud analysis covered all 73,000+ orders in the dataset. Every observed payment dispute, representing about 35,000 EUR, came from an order rated LOW risk.

That is not a bug in Shopify. Checkout fraud tools score the information available when an order is placed. Refunds, delivery claims, repeat orders and chargebacks happen later, so they require post-purchase history that a checkout-time score cannot yet see.

#shopify-fraud-analysis

75% of shipped-refund value went to customers with one observed order

Three quarters of refunded value went to customers with no later order in the dataset

About 900 customers placed one observed order, received a refund, and placed no later order during the dataset window. Together those refunds represented about 350,000 EUR, three quarters of all post-shipment refund value.

This does not establish fraud, acquisition cost, or lifetime value beyond the observed window. It shows where refunded value concentrated and why refund history belongs beside repeat-order behavior rather than in an isolated transaction list.

#cac-double-loss

1 in 6 customers who filed a chargeback came back and ordered again

Customers who filed a chargeback kept ordering, and the store kept shipping

About 1 in 6 customers who filed a chargeback returned and placed more orders. The store kept shipping because nothing in the default stack flags a past-chargeback customer at order time.

Separately, about a quarter of chargeback customers received another refund after their dispute. The source data does not establish that these were the same customers who reordered, so the two outcomes should not be read as a funnel.

#chargeback-repeat-customers

64% of bank disputes claimed the package never arrived

“It never arrived” was the dominant dispute reason, and most claims were checkable

64% of observed bank disputes were item-not-received claims, worth about 20,000 EUR. 84% of those claims had a carrier tracking number that could have been checked against delivery confirmation.

A tracking lookup at dispute time can produce evidence for these claims, either confirming a real delivery failure or contradicting the claim. Shopify's chargeback process specifically asks merchants for tracking, delivery confirmation, and customer communication.

#item-not-received

100% of observed shipped orders were refunded for a small repeat cohort

A small repeat cohort had every observed shipped order refunded

A small cohort with two or more shipped orders had every observed order refunded. Per order, each transaction looks like a normal purchase followed by a normal refund.

The dataset does not reliably record return receipt or product condition, so this is a review pattern, not proof that merchandise was kept or that fraud occurred. It only appears when refund history is rolled up to the customer level.

#free-shoppers

~2% of refund value came from the top 10 refunders

Refund exposure was diffuse, not concentrated

In this dataset the top 10 refunders accounted for about 2% of refund value, and customers with 3 or more refunds for less than 2%. Refunded value was spread across nearly 2,000 customers, most of them with one observed refund.

Repeat behavior still carries signal. More than 100 customers who refunded twice or more averaged about 690 EUR of lifetime refunds against 420 EUR for one-time refunders, roughly 63% higher. And when a customer's second refund came, it came fast: the median gap between the first two refunds was about 2 days. Repeat refunding arrives in bursts, not as a slow habit.

Rolled up to the customer level, refunds belonging to customers our analysis flagged MEDIUM risk or higher came to about 170,000 EUR, roughly 1 in 5 refund euros.

#diffuse-not-concentrated

2.7x higher refund probability on 700+ EUR orders vs sub-100 EUR orders

Refund probability climbs with basket size

Orders under 100 EUR refunded at 1.5%. Orders over 700 EUR refunded at 4.0%, a 2.7x gradient. Bigger baskets carry disproportionate refund exposure, which matters when deciding which orders deserve a manual look before fulfillment.

#order-value-gradient

~30% relative refund-rate lift on selected promotional-code orders

Discounted orders refunded more often

About 200 refunds sat on abnormally discounted orders, worth roughly 70,000 EUR. Orders placed with selected promotional codes refunded at 3% to 4% against a baseline of about 3%, up to roughly 30% higher in relative terms.

About 20% of refunds were for the full order value, totalling roughly 220,000 EUR.

#discount-stacked

Findings are not the same as money saved

Historical patterns show exposure. They do not prove what RefundSentry prevented. After installation, the product keeps a value ledger with three claims that never get blended into one inflated ROI number:

Verified recovery
Recorded only when the dispute has a merchant or system prefill/submission on record and the card network confirms it was won.
Observed avoided exposure
Recorded separately when RefundSentry holds an order and Shopify later confirms it was cancelled before fulfillment. It is exposure, not guaranteed savings.
Influenced activity
Completed actions can be counted, but flags, clicks and recommendations never become money by themselves.

Every monetary entry keeps its evidence, currency and correction history. The ledger stores no raw customer personal information.

Smaller findings worth knowing

One-third
of shipped-refund value went out in December and January. Returns season is real, and it is concentrated.
~3x lower
Multi-variant orders refunded below 1% against a baseline of about 3%. In this catalog, buying variants correlated with lower refund risk, the opposite of the usual bracketing assumption.
0.6%
Bracketing proper (ordering multiple variants to keep one) represented well under 1% of refunds. A defining pattern in some categories, but marginal in this dataset. Fraud patterns are vertical-specific.
~4x
The highest-refund product approached 6% of orders, while the lowest stayed below 2%. Product-level refund rates varied far more than customer geography or email domain ever did.
90+ days
A small cohort of refunds was issued more than 90 days after the order, worth over 10,000 EUR. Only a minority was tied to a chargeback.

A second store: what held and what did not

In September 2026 we ran the same analysis on a second anonymized Shopify store: about 48,000 orders and 2,800 refunds over 12 months, roughly 850,000 USD refunded, US-based with an average order around 280 USD, in a different vertical from store A. Same method, same rounding, same denominators. Store B had too few bank disputes (15) to compare, so the dispute finding below is a store A update.

Held in both stores

42%

The never-shipped half is a two-store pattern

42% of store B's refunded value (about 350,000 USD) sat on orders cancelled before fulfillment, against 46% in store A. A refund dashboard that mixes the two halves overstates post-shipment exposure by close to 2x in both catalogs.

Held in both stores

2.3 days

The second refund still comes fast

Counting refunded orders rather than refund lines, 115 customers in store B had two or more refunded orders. The median gap between the first two was 2.3 days (about 2 days in store A). 62% arrived within a week and 84% within 30 days. Repeat refunding is a burst in both stores.

Held in both stores

4x

Repeat refunders cost more per head

Customers with two or more refunded orders averaged about 1,370 USD of refunds against 340 USD for one-time refunders. They were 5% of refunding customers and held 19% of refund value. The 31 customers with three or more refunded orders held 8%.

Did not hold

6.7%

Concentration was three times higher, and still diffuse

The top 10 refunders held 6.7% of refund value in store B against about 2% in store A. More concentrated, but 93% of the money still sat outside the top 10. A blocklist of the worst offenders would not have moved either store's refund line.

Did not hold

2.8x

Shared addresses refunded more, this time

Orders shipped to an address already used by another customer account refunded at 8.9% against 3.2% for single-account addresses. Addresses shared by three or more accounts refunded at 16%. That is 3.4% of orders and 1.5% of addresses (about 800), a list small enough to review by hand. In store A the same clusters were families and pickup points and refunded at baseline. The pattern is real and store-specific, which is why it has to be measured per store rather than assumed.

New

23x

Disposable email addresses were the strongest single tell

53 orders in store B were placed with a disposable email address. 42 of them were refunded: a 79% refund rate against a 3.4% baseline. Store A had too few to measure. Small count, very large effect, and it is visible before the order ships.

Held in both stores

2.7x

Basket size still predicted refunds

Orders flagged as a high-value first order refunded at 9.0% against 3.4%. Store A showed a 2.7x gradient between sub-100 EUR and 700+ EUR orders. Two stores, two currencies, same slope.

New

1 in 3

Store A update: a third of the never-arrived claims had a delivered scan

Re-running store A on the 12 months to September 2026: 48 item-not-received disputes, 40 with a tracking number and 17 with a carrier delivered scan on record. The merchant won 28 of the 48. The delivery evidence existed for a third of the claims.

What we checked and did not find

Negative results rarely get published, which is how the same myths keep circulating. These angles were tested on the same dataset and came back flat:

  • Shared-address clusters barely refunded. The biggest multi-account addresses turned out to be families and pickup points, not rings.
  • Geography was flat. No region refunded meaningfully above baseline.
  • Email domain told us nothing: mainstream and custom-domain customers refunded at effectively the same rate.
  • Guest checkout was a non-factor in this dataset.
  • Refunds did not cluster against the return-window deadline. More than half landed within 7 days of the order.

One caveat on scope: this is a single store in a single vertical. The bracketing and multi-variant numbers in particular would look different on an apparel store, and the null results above are store-level observations, not category laws.

Methodology

Source. Production data from one anonymized Shopify store, analyzed with permission. Window: 12 months in 2025–2026. Public base: more than 73,000 orders, 2,100 refunds, about 860,000 EUR refunded, nearly 2,000 refunding customers and roughly 70 bank disputes.

Second store (September 2026). Store B is a second anonymized Shopify store analyzed with the same method over 12 months to September 2026: about 48,000 orders, 2,800 refunds, roughly 850,000 USD refunded, about 2,100 refunding customers and 15 bank disputes. Refund rates for store B use orders older than 45 days as the denominator, exclude orders with a bank dispute, and count a refund only when the order was not cancelled. Repeat-refunder figures count refunded orders, not refund lines, because partial refunds on one order would otherwise inflate them. The shared-address figure comes from a privacy-preserving address fingerprint (a one-way hash of the normalized address), never from the address text.

Shipped vs cancelled. Refunds were split by whether the underlying order was ever fulfilled. About 800 refunds (roughly 400,000 EUR) sat on never-shipped orders; about 1,300 refunds (roughly 460,000 EUR) on shipped orders. Post-shipment statistics use the shipped slice as denominator because a refund on a cancelled order cannot be return abuse.

Wording.We say a customer "filed a chargeback", not "is a fraudster". Risk flags describe statistical patterns, not verdicts about people, and observed behavior does not establish motive.

Limitations.Single store, single vertical, single 12-month window. Lifetime value beyond the window, customer-acquisition cost, return receipt and product condition were not measured reliably. Refund-method splits were excluded because a third of refund value sat in an unexplained "other" bucket. Public counts and amounts are rounded for privacy; percentages and ratios were calculated from full-precision source data, so displayed figures may not recompute exactly. No row-level data is published.

Citing this page.Free to cite with attribution: "RefundSentry return fraud research, 2026", linked to https://refundsentry.com/research/return-fraud-statistics. Licensed CC BY 4.0.

Frequently asked questions

Where does this return fraud data come from?
From 12 months of production data in 2025-2026 for one anonymized Shopify store: more than 73,000 orders, 2,100 refunds, about 860,000 EUR of refunded value, nearly 2,000 refunding customers and roughly 70 bank disputes. In September 2026 a second anonymized store (about 48,000 orders and 850,000 USD refunded) was run through the same method. RefundSentry ran both analyses with permission. Published counts and amounts are rounded; percentages and ratios use the full-precision source data.
What share of refunds is actually fraud or abuse?
This dataset cannot establish an exact fraud share. 46% of refunded value was on orders that never shipped (mostly formal cancellations), so it was not return abuse. On the shipped half, the top 10 refunders held about 2% of refund value, while about 1 in 5 refund euros belonged to customers flagged as elevated risk. A risk flag is a review signal, not confirmation of fraud.
Why didn't Shopify's built-in fraud analysis catch these patterns?
Because the relevant events happened later. In this dataset Shopify's fraud analysis rated every order that later became a payment dispute as low risk. Refunds, item-not-received claims, repeat orders and chargebacks require post-purchase history that was not available when the checkout-time score was calculated.
Can a refunded Shopify order still become a chargeback?
Yes. A refund and a card dispute are separate events, so a customer can receive a refund and still dispute the original payment. Merchants need to connect refund history, later orders and chargeback outcomes at the customer level to catch that sequence before another order ships.
Can I cite these statistics?
Yes. Cite them as “RefundSentry return fraud research, 2026” with a link to this page. The source store and individual customers are not identified, and public counts and amounts are rounded. If you need the methodology behind a specific number, the definitions are on this page; for anything else, contact us.

Last reviewed: 1 September 2026. Disclosure: RefundSentry sells fraud-intelligence software for Shopify, so we have skin in the game on the detection framing. The numbers above stand on their own.

Go deeper

See what a year of your refunds is hiding

Free 12-month audit, scored report the same day, and 14 days to act on it before you pay.

Start Free 14-Day Trialor see an example report first →
Free 12-month auditPlans from $49/moCancel during the trial, pay $0