Velocity
How many attempts come from a single card, device, or IP, the fingerprint of a card-testing run.
AI fraud prevention
AI-Powered Fraud Prevention
We screen every transaction with AI risk scoring before it posts: velocity, device, and behavioral signals plus known card-testing and stolen-card patterns, so fraud is stopped before it becomes a chargeback. Tuned to each vertical to keep false positives low.
- Real-time scoring before posting
- Tuned per vertical
- AI screens, humans decide
Why it leads
The cheapest chargeback is the one that never happens
Fraud prevention is the part of payments most processors treat as an afterthought and we treat as the headline. The reason is simple arithmetic. When a stolen card slips through, you lose twice: once on the fraudulent transaction itself, and again on the chargeback it generates, which also pushes your dispute ratio toward the card-brand thresholds that can terminate an account.
So we screen every transaction in real time, before the payment posts. The model weighs the signals a static rule would miss and scores the transaction in the moment: clean ones go through instantly, risky ones get held or declined before they cost you anything.
The signals
What the model actually reads
Device & behavioral signals
Whether the session looks like a returning human or an automated script.
Billing-to-shipping mismatch
Address and CVV verification, the classic tells of a stolen card shipped elsewhere.
Known attack patterns
Matches against card-testing and stolen-card patterns seen across accounts.
Calibration
Tuned to your vertical, not a generic template
A screening model that's too aggressive is its own problem. Every legitimate customer it blocks is a lost sale, and a generic one-size template will flag your ordinary buyers as suspicious because it doesn't know what normal looks like in your business.
So the rules are calibrated to each vertical's real patterns. A subscription renewal, a large furniture order, a phone-keyed B2B payment, a supplement reorder: each has a different normal, and the model is tuned to those. The goal is a low false-positive rate. Stop the fraud, let the real customers through instantly.
The full toolkit
The controls behind the score
Real-time AI scoring is the brain, but it works through a set of concrete controls, the explicit capabilities a high-risk account needs to keep fraud and friendly fraud off the books.
- Real-time AI risk scoring on every transaction before it posts, weighing velocity, device, and behavioral signals.
- 3-D Secure 2.0 / SCA authentication, applied to the transactions that warrant it, shifting fraud-chargeback liability to the issuer where it qualifies.
- Network tokenization that keeps real card numbers off your systems and auto-updates reissued cards, cutting both risk and involuntary churn.
- AVS and CVV verification to catch billing-to-shipping mismatches and the classic tells of a stolen card.
- Clear and dynamic billing descriptors that prevent the friendly-fraud disputes caused by an unrecognized charge on the statement.
- Velocity and rules tuned per vertical, calibrated to your business's normal so legitimate customers aren't flagged.
Two problems
Fraud and chargebacks are different problems
| Fraud | Chargeback | |
|---|---|---|
| What it is | An illegitimate transaction: stolen card, takeover, testing run | A reversal, sometimes true fraud, often friendly fraud |
| When it's caught | Stopped before it posts | Defended after a dispute is filed |
| The defense | Fraud prevention (this page) | Chargeback management |
Risk that gets worked
Flagged transactions surface with the full context
Every payment is scored on velocity, device, location and behavior, and flagged the moment the signals don't add up. When something is off, the transaction is held and surfaced with its cause and context for review, not waved through to become a chargeback later.
- Velocity, device and location signals
- Declines and holds with the full context
- Tuned per vertical to keep false positives low
FAQ
Fraud prevention FAQ
How does AI prevent payment fraud?
It scores every transaction in real time before the payment posts, weighing signals a human or a static rule would miss: velocity (how many attempts from a card, device, or IP), device and behavioral fingerprints, billing-to-shipping mismatches, and patterns that match known card-testing or stolen-card attacks. Transactions that score clean go through instantly; risky ones get held or declined before they settle. The advantage of AI over fixed rules is that it adapts to new fraud patterns instead of only catching the ones someone already wrote a rule for.
How do high-risk merchants stop chargebacks?
The cheapest chargeback is the one that never happens, and most fraud chargebacks can be stopped before fulfillment. Screening transactions for stolen-card and card-testing patterns before they ship keeps fraudulent orders off your books and off your chargeback ratio. Pair that pre-transaction screening with dispute representment for the chargebacks that do come through, plus the prevention basics (clear descriptors, honest delivery terms, fast customer service), and you keep your ratio under the card-brand thresholds that put accounts at risk.
What's the difference between fraud and a chargeback?
Fraud is an illegitimate transaction: a stolen card, a card-testing attempt, an account takeover. A chargeback is a customer (or their bank) reversing a charge, which can be true fraud but is often 'friendly fraud': a real customer disputing a charge they actually made. Fraud prevention stops the first kind before it posts; chargeback management defends against both kinds after a dispute is filed. You need both, and they work together.
Does fraud screening slow down or block real customers?
Good screening is tuned to let legitimate transactions through instantly while catching the risky ones. The goal is a low false-positive rate, because blocking real customers costs you sales just as fraud costs you money. The model scores in real time, so a clean transaction isn't delayed, and the rules are calibrated to your vertical's normal patterns rather than a generic template, which is what keeps it from flagging your ordinary customers as suspicious.
Do you support 3-D Secure (3DS 2.0) and SCA?
Yes. 3-D Secure 2.0 adds an authentication step that verifies the cardholder with the issuing bank at checkout, and where it applies it shifts liability for fraudulent chargebacks from you to the issuer. It's part of the toolkit alongside AI risk scoring: the model decides when a transaction warrants a 3DS challenge so you authenticate the risky ones without adding friction to the clean majority. For merchants who need Strong Customer Authentication, 3DS is how that requirement is met within the same flow.
What is network tokenization and how does it help?
Network tokenization replaces the card number with a token issued by the card networks themselves, which updates automatically when a card is reissued or expires. That does two things for fraud and revenue: it keeps real card numbers out of your systems (shrinking both risk and PCI scope), and it keeps stored cards current so recurring charges don't fail when a customer's card is replaced. It works hand in hand with our tokenization and recurring billing, protecting card data and reducing involuntary churn at the same time.
Related
Keep exploring
Stop fraud before it becomes a chargeback.
Fraud prevention isn't sold as a separate product here, it's built into every account, tuned to your vertical at onboarding.