Shopify Fraud Analysis: How the Machine Learning Model Works
Shopify fraud analysis uses machine learning trained on historical transactions across all Shopify stores to give eligible online card orders a low, medium or high risk of a fraud chargeback. Next to that recommendation, the order shows individual fraud indicators, such as whether the card passed address and CVV checks. Act on the recommendation; a single red indicator does not decide the order on its own.
Sources, read on October 23, 2026: the Shopify Help Center page on fraud analysis and its fraud prevention overview, and Shopify’s blog post on its fraud technology of 2023, whose figures are Shopify’s own.
What the model looks at
Shopify describes its model as “supervised machine learning”, an adaptive system that evaluates the risk of each payment in real time. In 2023 Shopify wrote that it had trained it on over 10 billion transactions across millions of merchants. The signals it names include:
- Country and IP address
- Customer behavioral patterns
- Payment details and wallet type
- Past customers and encounters already flagged as fraudulent
The Help Center adds that the algorithms are continuously improved. As a merchant you see its output on each order: a recommendation and a list of indicators.
Indicators vs the recommendation
On the order details page, the Order risk section shows the analysis. Depending on the view it is labeled Order risk evaluation or About this order.
Fraud indicators are individual facts about the order. They can include:
- whether the card passed Address Verification System (AVS) checks
- whether the customer entered the correct CVV
- whether the customer’s location matches the payment method
- unusual device or network activity
- whether the customer tried more than one card
Indicators are marked as associated with fraudulent behavior, legitimate behavior, or neutral; some views group them under High risk signals, Low risk signals and Other details. The IP address and a location description are listed separately, and on their own they are not a fraud recommendation.
The fraud recommendation is the overall judgment, and it comes with guidance: You can fulfill this order, Confirm with the customer before fulfilling, or Consider canceling this order. Shopify is explicit that individual indicators and group headings do not represent the order’s overall risk level. An order can show one red indicator, such as an address check that failed, and still be low risk overall.
What to do at each risk level
| Recommendation | What Shopify shows | Your action | Automate with Flow |
|---|---|---|---|
| Low | No warning | Fulfill as normal | Nothing needed |
| Medium | Warning symbol next to the order number | Look at the indicators; contact the customer if something does not fit, then fulfill | Tag the order and email staff to review |
| High | Warning symbol; guidance usually says to consider canceling | Verify with the customer, or cancel and refund before fulfilling | Hold fulfillment and notify you |
Flagged orders are also highlighted in order notification emails if you subscribe to them. If you want time to review before the payment is taken, set up manual payment capture. Settings for these steps are in our guide to Shopify fraud settings, and the automation logic in the Shopify Flow guide.
Who gets indicators and who gets recommendations
- Fraud indicators and support for third-party fraud apps: Basic plan or higher.
- Fraud recommendations: Grow plan or higher, or any plan with Shopify Payments.
- Analysis works on online credit card orders that Shopify can verify. On a third-party payment processor, ask the processor about its own limits.
Shopify Payments can also block some high-risk checkout attempts before an order exists. Those never become orders, so they have no recommendation; you find them in the Blocked view of Orders > Abandoned checkouts. If you recognize a blocked customer, create a draft order and send them an invoice.
Orders without a score
These receive no fraud recommendation: test orders, free orders, orders paid entirely with a gift card, Point of Sale orders, B2B orders and subscription renewals. You can simulate a recommendation on a test order to see the view. Treat a missing score as unchecked and use your own judgment on large orders in these groups.
When a real customer looks risky
Merchants on the Shopify Community report real customers flagged as high risk when they shop through a VPN. A customer traveling abroad, or retrying with a second card after a decline, touches indicators on Shopify’s own list too: location against payment method, and more than one card. The way out is the one Shopify recommends for any medium- or high-risk order: contact the customer before fulfilling. In the desktop view, Contact customer opens an email to them. A genuine buyer answers quickly and knows the details they used.
Do not judge by the IP location alone; Shopify says an IP address or location description by itself is not a fraud recommendation.
The tools around the model, such as Shopify Protect, card testing protection and 3D Secure, are covered in our guide to Shopify fraud prevention, with the rest of our running a store topics.
Frequently asked questions
How accurate is Shopify fraud analysis? Shopify does not publish an accuracy rate for merchants. It says the model is trained on historical transactions across all Shopify stores and continuously improved; in 2023 it reported training on over 10 billion transactions. Use the recommendation as a strong signal and check medium- and high-risk orders yourself.
Can I change an order’s fraud risk level? You cannot edit Shopify’s recommendation. You decide what to do with the order: fulfill it, contact the customer, or cancel and refund. Third-party fraud apps, supported from the Basic plan, can add their own assessments; when an order has several, you choose which provider’s assessment to view.
Does Shopify cancel high-risk orders automatically? Shopify does not cancel orders on the recommendation by itself, although Shopify Payments can block some high-risk checkout attempts before they become orders. To act automatically on high-risk orders, build a Shopify Flow workflow that holds or cancels them.