Where AI Fits in a Mature Fraud Program

The strongest fraud programs use AI to sharpen detection, reduce friction, and uncover connections across accounts, devices, identities, and transactions—while keeping people accountable for consequential decisions. That combination of automation, layered controls, governance, and human judgment is becoming increasingly important as fraudsters use the same technology to make attacks more convincing, easier to scale, and increasingly organized like business operations. 

In a ProSight Banking Strategies podcast discussion, Bobbie Paul, managing director for fraud at Huron, described this approach as “accountable augmentation.” Banks have access to strong data, behavioral analytics, fraud intelligence, and established controls, she explained, but they also must ensure that AI-supported decisions remain accurate, explainable, and subject to meaningful challenge. 

Paul emphasized several priorities for putting that approach into practice: 

Build layers rather than relying on one model. Effective fraud programs combine behavioral analytics, anomaly detection, entity resolution, link analysis, supervised machine learning, unsupervised pattern detection, and other controls. Graph and network analytics can be especially valuable because fraud often connects identities, devices, accounts, and transactions that appear unrelated when viewed separately. 

Institutions should deploy those capabilities according to risk rather than applying every control everywhere. The objective is to strengthen defenses without creating unnecessary cost or making every legitimate interaction difficult. 

Use AI before and after the transaction. Real-time tools can help stop fraud before money leaves the institution or an account is taken over. Post-transaction analysis remains equally useful. It can uncover emerging tactics, improve models, identify organized networks, and reveal whether an incident was isolated or part of a larger campaign. 

That broader view matters because suspicious activity is not always apparent at authorization—and banks cannot insert friction into every transaction. 

Make protective friction understandable. AI should help reduce false positives, but Paul cautioned against treating every security interaction as a customer experience failure. Text verifications, analyst calls, and other checkpoints can reinforce trust and brand loyalty when customers understand that the bank is protecting their money and identity. 

Govern the entire AI lifecycle. Banks should be able to explain where AI is used, what data feeds it, who can modify or challenge a model, which attributes influence decisions, and how performance is monitored over time. Model drift, bias, weak data, and automation complacency all create risk when oversight is insufficient. 

AI governance should also be integrated into the fraud risk assessment rather than added later as a separate exercise. 

Keep people responsible for consequential decisions. Experienced investigators and risk professionals remain essential for validating outputs, spotting patterns, and challenging assumptions. Paul expects human oversight to remain particularly important for high-impact actions such as closing an account, denying access to funds, or terminating an employee. 

The takeaway: As Paul put it, “The future of AI and risk is accountable augmentation.” The strongest programs will pair smarter automation with clear governance, layered defenses, and people who remain empowered to question what the technology produces.