For years, financial institutions have invested heavily in AI-powered fraud detection. And near daily headlines reinforce its importance with fraud and scams exploding around the world. Those investments are essential, particularly as real-time payments (RTP), FedNow®, P2P platforms, account-to-account transfers, and digital wallets continue to grow. But as fraud evolves, an uncomfortable reality remains:

Not every fraudulent, disputed, or questionable transaction can be stopped before it happens.

Even the most sophisticated fraud models must balance risk against customer experience. False-positive rates remain uncomfortably high. Block too much, and legitimate transactions are declined. Block too little, and fraud slips through. A recent survey estimated “61.3 million Americans had fraudulent charges on their credit or debit cards in the past year, totaling roughly $6.1 billion in unauthorized purchases.”

Additionally, while most victims noticed the charges rapidly – aided by alerts from their bank – roughly 38–46% didn’t notice until later. And 10% didn’t catch it for one or more billing cycles. However, alerts may not be as effective as we believe. A 2026 webinar by Javelin on their research found "55% of consumers took no action on a fraud alert because of scam suspicions." This is where AI detection alone is not enough.

As a result, you are facing not just higher volumes today of claims that were not stopped, but an impending tidal wave of volume that must be handled after the transaction posts.

The growing operational challenge

In an increasingly fragmented payments landscape, real-time payments provide limited opportunities to stop a transaction once funds leave an account. Fraudsters understand this. At the same time, fraud schemes are increasingly blended with social engineering, authorized push payment fraud, first-party misuse, and merchant-related disputes.

The result is a growing backlog of cases requiring investigation, regulatory evaluation, evidence collection, network interactions, and customer communications to reach resolution. Many of these processes remain highly deterministic. Network regulations from Visa and Mastercard prescribe specific timelines, reason codes, evidence requirements, and processing steps. Consumer protection regulations, such as Regulation E and Regulation Z, similarly define obligations regarding investigations, provisional credits, customer communications, and resolution deadlines. If the outcome requires adherence to documented rules, AI should not be making unconstrained decisions.

The right AI for the right decision

The most effective dispute operations use multiple forms of AI – not a single large language model everywhere.

For example:

• Predictive AI and machine learning can identify disputes likely to succeed, predict representments, prioritize work, and recommend optimal treatment paths.
• Generative AI can summarize customer interactions, explain policies, guide employees, accelerate knowledge retrieval, and read unstructured documents.
• Deterministic business rules and workflow automation ensure compliance with regulatory and network requirements while delivering consistent outcomes.

This layered approach allows institutions to apply AI where judgment, prediction, or content generation adds value – all while preserving auditable, prescribed processing where regulations demand consistency.

Minimizing cost while maximizing trust

As organizations rush toward agentic AI, many are discovering that indiscriminate use of generative AI introduces unnecessary cost, inconsistency, and governance challenges.

The better approach is to reserve token-intensive AI interactions for specific moments where they create measurable value – summarization, employee assistance, customer guidance, or knowledge retrieval – while allowing deterministic automation and workflow orchestration to manage the majority of case processing.

In an era of increasing fraud and dispute volumes coupled with payment complexity, success will not belong to institutions with the best fraud model alone. It will belong to those that combine AI, workflow automation, compliance, and human expertise into the best operating model.

Because the true challenge is no longer just detecting fraud.

It is managing everything that happens after detection – and doing so at enterprise scale, with speed, consistency, compliance, and trust.

Automation is the real force multiplier

The future of dispute management is not autonomous AI working in isolation. Modern dispute operations require orchestration of AI, workflows, decisions, systems, and humans working together. Straight-through processing can automate qualifying cases, while workflow automation eliminates repetitive activities, such as case routing, system updates, accounting entries, evidence collection, customer notifications, and network submissions.

More importantly, workflow provides a governance layer around AI.

Every recommendation can be reviewed, every action audited, every decision traced back to a rule, policy, regulation, or model. Human experts remain in the loop where exceptions arise or judgment is required, while automation handles the volume that would otherwise overwhelm operations teams.

Pega Smart Dispute Agentic Automation combines AI-powered decisioning with workflow automation to manage the full lifecycle of fraud claims and payment disputes across cards, ACH, RTP, P2P, and emerging payment types – in any channel. The focus here is orchestrating what happens next when fraud isn’t prevented. With support for payment network rules and government regulations, Smart Dispute applies the right AI under the right circumstances to maximize efficiency, achieve service-level obligations, and drive customer satisfaction.