Modern fraud is adaptive, relentless, and increasingly automated. It duplicates itself faster than traditional detection systems can respond. Financial institutions face adversaries who use the same AI tools defenders rely on, creating an arms race where speed and accuracy determine who wins.
David Fapohunda has spent over two decades leading operations, fraud strategy, and AI-driven transformation across financial services. From policy design to platform integration, he has built systems that process millions of transactions per hour while maintaining the precision needed to catch sophisticated threats before they cause damage.
“AI can not only detect fraud faster, but it can also predict and prevent it before it happens,” Fapohunda explains. “But scaling that capability requires more than deploying algorithms. It requires the right infrastructure, data strategy, and human intelligence working together.”
Build on Data, Not Just Algorithms
AI is only as strong as the data behind it. Fraud detection models fail when they operate on incomplete information or siloed data sources. Scaling starts with integrating diverse, high-quality data across the enterprise, both structured and unstructured, from every system that touches customer activity.
During major transformation programs, Fapohunda has invested early in unifying data pipelines across contact centers, transaction systems, and third-party platforms. This gives models the full context they need to spot anomalies that would otherwise remain invisible.
“We integrated data from channels that had never been connected before,” says Fapohunda. “Once the models had visibility across the entire customer journey, detection accuracy improved significantly and false positives dropped because the system understood normal behavior patterns in context.”
This delivers measurable outcomes. Unified data pipelines reduce the time it takes to identify emerging fraud typologies from weeks to days. Models trained on comprehensive datasets catch threats earlier in the attack chain, preventing losses rather than reacting to them. Detection rates improve without increasing friction for legitimate customers because the system has enough context to distinguish between suspicious activity and unusual but valid transactions.
Pair AI With Human-Informed Intelligence
AI can detect patterns, but humans understand intent. The best fraud programs use AI to surface high-risk signals, then leverage seasoned investigators to validate those signals and provide the nuanced judgment machines cannot replicate.
Fapohunda has built feedback loops where investigators train models over time. This improves accuracy and minimizes false positives. When AI flags activity and investigators determine it was legitimate, that insight feeds back into the model so it learns to distinguish between similar patterns in the future.
“We use digital investigation engines to triage, review, and risk-map workflows,” Fapohunda explains. “For more complex cases, seasoned experts vet and verify high-risk events. That combination gives us both speed and precision.”
Human-informed AI reduces investigator workload by filtering out low-confidence alerts, allowing teams to focus on the cases that require deep expertise. It also accelerates model improvement because every investigated case becomes training data that sharpens future detection.
Design for Scale From Day One
If a fraud system cannot keep up with volume, it becomes a bottleneck. Scalable fraud detection requires cloud-native infrastructure, real-time analytics, and modular AI models that can adapt to new fraud typologies as they emerge.
Fapohunda has implemented service-oriented architectures that support millions of transactions per hour. The key differentiator was leveraging large graph databases to provide fuller context windows of device, customer, and counterparty behavior.
“Graph databases let us map relationships across entities in ways traditional systems cannot,” says Fapohunda. “We can see when a device connects to multiple accounts, when accounts share unusual behavioral patterns, or when counterparty networks indicate coordinated fraud attempts. That visibility is critical when you are processing volume at scale.”
Designing for scale from day one means fraud detection capabilities grow with transaction volume without requiring platform rewrites. Real-time analytics ensure threats are identified and stopped before transactions clear. Modular models allow teams to deploy new detection logic for emerging threats without disrupting existing capabilities.
“Fraud is not standing still, and neither should your defenses,” Fapohunda concludes. “By aligning the right data, human expertise, and scalable AI infrastructure, you can move faster than the threats. Build fraud resilience that learns, adapts, and protects at scale.”
Connect with David Fapohunda on LinkedIn for insights on scaling AI-driven fraud detection capabilities.
