1. Home
  2. Blog
  3. Payments

E-commerce fraud prevention: approve more sales, stop more fraud

Learn how e-commerce fraud prevention works, from rules and risk scores to 3DS and manual review, and how to tune it to approve more legitimate orders safely.

Payments

E-commerce fraud prevention is the set of rules, models and checks that decides, within seconds, whether an order looks legitimate before the payment goes on for authorization. The challenge cuts both ways: stopping people who use someone else's card without turning away good customers, because a declined legitimate sale is expensive too. A well-tuned fraud system watches both errors at once.

The two errors that cost money

Every fraud system makes two kinds of mistakes. It can approve a fraudulent order, which turns into a chargeback weeks later, or it can decline a legitimate customer, known as a false positive.

ErrorWhat happensVisible costHidden cost
Fraud approvedChargeback for an unrecognized purchaseSale amount, goods and feeExposure to card network monitoring programs
False positiveGood customer declinedNone at the timeLost sale and a customer who never returns

The trouble with false positives is that nobody complains about them. The declined customer simply buys from a competitor. An overly strict system keeps the chargeback dashboard looking clean while revenue leaks out without showing up in any report.

Layers of protection

An online order passes several checkpoints before approval. Each one sees a different slice of the risk:

  • Your site or app: attempt limits, email and phone verification at signup and bot protection.
  • Fraud screening: the service from your gateway, payment intermediary or a specialist vendor, which returns a risk score or a decision for each order.
  • 3DS authentication: the issuer confirms that the buyer is the cardholder, often without friction and, when needed, with a challenge in the banking app.
  • Issuer analysis: the card's bank applies its own models before approving.
  • Manual review: a person analyzes orders that land in the gray zone.

3DS has an important effect beyond security. In many cases, authenticated transactions shift liability for fraud chargebacks to the issuer, according to each card network's rules. That makes it a good extra step for medium-risk orders instead of an outright decline.

Rules, scores and manual review

Rules are if-this-then-that instructions. For example: block more than three different cards on the same device within an hour, or send for review a high-value order from a freshly created email with express shipping to an address different from the account. Rules are transparent and easy to adjust, but they are rigid, and fraudsters learn to work around them.

A risk score is a number calculated by statistical models that weigh dozens of signals at once. It catches patterns a rule would miss, but it depends on data volume and on receiving feedback about which orders turned out to be fraud.

Manual review steps in where the machine is unsure. It is slow and costly, so it makes sense for high-value orders in the middle risk band, with a set response time so deliveries are not delayed.

The setup that usually works best combines all three: rules for the obvious cases, scoring for volume and manual review for the gray zone, with 3DS as a confirmation step.

Signals that separate customers from fraudsters

No single signal proves fraud. What counts is the combination:

  • Device and browser: the same device used with several identities or cards.
  • IP and location: a large distance between the connection, the billing address and the shipping address.
  • Email and phone: accounts created minutes before the purchase or with no history.
  • Data consistency: a name, tax ID and cardholder that do not match. In Brazil, the individual tax ID is the CPF.
  • Store history: earlier paid, undisputed purchases sharply reduce risk.
  • Session behavior: every field pasted, straight to expensive items, unusual speed.
  • Order profile: easy-to-resell items, several identical units and express shipping.

Collecting this data has to follow privacy law. In Brazil that means the LGPD, Law 13.709/2018. Fraud prevention can justify the processing, but the purpose must be described in your privacy policy and the data must be protected.

Card testing and other attacks

Some attacks are not after a purchase at all, but after validating stolen data. In card testing, bots place dozens of low-value orders to find out which cards work. The signs are a run of declines, small amounts and many attempts from the same IP range or device. Defenses include attempt limits per IP, device, email and card, bot checks after declines and blocking checkout after a set number of failures.

Account takeover is another common attack. The fraudster logs in with a leaked password, changes the address and buys with the saved card. Asking for a confirmation code on sensitive changes and alerting customers about logins from new devices reduces the risk.

With PIX, Brazil's instant payment system created and run by the Banco Central do Brasil, the risk looks different. There is no card fraud, but there are social engineering scams, fake payment receipts and refund requests to different accounts. There, fraud prevention focuses on the customer account and the refund flow.

How to tune and monitor

Fraud prevention is not set-and-forget. Track the overall approval rate, the fraud screening decline rate, fraud chargebacks by count and value, the manual review rate and review turnaround time every month.

  1. Define your risk appetite by product and channel, since an event ticket and a smartphone have different profiles.
  2. Feed every fraud chargeback back into the system so the model learns.
  3. Review rules often and remove those that only block good customers.
  4. Test changes on part of your traffic before rolling them out everywhere.
  5. Estimate false positives by reviewing a sample of declined orders.
  6. When you decline, offer another payment method instead of blocking the customer outright.

Bridgemoon has its own payment processing system and builds the integrations that feed checkout and app signals into fraud screening, along with chargeback feedback to adjust the rules.

Frequently asked questions

Does 3DS eliminate chargeback risk?

No. In many cases, transactions authenticated with 3DS shift liability for fraud chargebacks to the issuer, according to each card network's rules. It does not protect against merchant disputes, such as items not received or ignored cancellations.

Does fraud screening slow down order approval?

Automated analysis takes a few seconds and customers barely notice it. Delays appear when an order goes to manual review, which can take hours. Reserve manual review for high-value cases and set a maximum response time.

Is manual order review worth it?

It is worth it for high-value orders in the middle risk band, where declining would mean losing a good sale. For low tickets, the cost of review usually exceeds the amount at stake. In that case, rules, scoring and 3DS do the job better.