Why Fraud Defense Is Becoming a Speed Problem, Not Just a Detection Problem

Part 2 of 5: Gradial FinServ Intelligence Series

Sixty-one million Americans had a fraudulent charge on their card in the past year, totaling roughly $6.1 billion in unauthorized purchases. But a more revealing shift is happening underneath that headline: first-party fraud, where a customer disputes a charge they actually made, has grown from 7.6% of cases in 2023 to 30.4% in 2024, now roughly matching third-party fraud in scale. Account takeover attempts are up 141% since 2021, and deepfakes, a category that barely existed three years ago, now account for 11% of fraud globally. Fraud is not simply growing; its shape is changing quickly enough that the tools built to catch yesterday’s patterns are increasingly working against a different problem.

This is less a modeling challenge than an operational one. Most large issuers have had fraud scoring in place for a decade, and the underlying models are reasonably strong. What’s under the most pressure is the human review layer that sits behind every flagged transaction: the investigation, documentation, and resolution work that still happens largely by hand, at a pace that struggles to keep up with how quickly fraud typologies are multiplying. Fraud and disputes have, as a result, become one of the largest line items in a card platform’s operating budget, with average losses of $60 million per organization in the past year, a meaningful share of which reflects the labor and false positives behind the review process itself.

The data suggests a real divide is opening between issuers who have moved AI into that workflow and those who haven’t: 42% of issuers and 26% of acquirers report saving more than $5 million over the past two years by applying AI directly to fraud and dispute resolution, and 83% say it has measurably reduced both false positives and customer churn. The difference isn’t the sophistication of the underlying fraud score. Most issuers have comparable scoring today. It’s whether that score can be acted on in real time, across systems, rather than routed into a queue for manual review.

It’s a useful moment for risk and fraud teams to ask where their organization sits on that divide, not because manual review is obsolete, but because the fraud landscape is evolving fast enough that the workflow behind the score is becoming as important as the score itself. A related version of this problem shows up on the content side for card issuers: terms, fraud-alert messaging, disclosures, and campaign pages need to be updated just as quickly and accurately as the fraud rules behind them, often across a site that’s just as fragmented as the review queues described above. That’s the layer Gradial’s agents operate in, pushing coordinated updates to disclosures, terms, and marketing pages across the CMS, with governance and audit trails built into execution rather than added after the fact.

Previously in Part 1: The AI Paradox in Asset Management: Widespread Adoption, Elusive Returns.

Up next in Part 3: what’s really separating AI leaders from laggards in insurance underwriting.

• Security.org: Credit Card Fraud Report
• The Financial Brand: The Credit Card AI Crime Wave, and How to Fight Back in 2026
• PYMNTS: Card Issuers Link Fraud Defense to Trust
• Mastercard: AI Is Helping Banks Save Millions by Transforming Payment Fraud Prevention
• Mastercard: Payments Fraud Is Growing in Scale and Sophistication
• Sumsub: Fraud Trends 2026: AI Scams, Deepfakes, and Emerging Threats