Financial crime detection is quietly being rebuilt from the ground up. What was once a rules-driven discipline defined by static thresholds, retrospective reviews, and ever-growing backlogs of alerts is now being reshaped by machine intelligence that can detect patterns, context, and intent in real time.
This change is not just technological. It’s a structural thing. Detection is moving from linear screening to dynamic, adaptive systems that continuously learn from data, behaviors, and risk signals across the enterprise.
For Scott Nice, CRO at Label, the difference is clear. AI excels when scale and complexity overwhelm traditional approaches.
“AI clearly outperforms rules when your job is to find patterns across large amounts of data that human teams, and frankly traditional rules engines, cannot process in a meaningful way,” he says. This is especially true in dynamic environments where behavior evolves over time, risk is spread across multiple weak signals, and where companies need to distinguish between truly suspicious activity and legitimate but anomalous customer behavior.
In his view, the rules are still useful, but they have their limits. “They are usually best at understanding what is already known.” AI, by contrast, is most powerful when the question moves from detection to discovery. “What am I missing? What am I doing wrong? What is worth looking into?”
But Nice is clear that human judgment remains important, especially at the point where detection becomes decision-making. “That’s not something companies should try to automate.”Assessing credibility, weighing evidence, and interpreting context remains essentially a human job.
He draws parallels with broader compliance functions, from tax to customer due diligence. Automation can handle data validation, identify discrepancies and surface risks, but “someone still needs to understand the context and support decisions, even when the facts are no longer clear and simple.”
The future of financial crime detection
Nice sees the trajectory of financial crime detection as undeniably network-driven.
“Financial crime does not respect the boundaries of a single institution, so the direction of movement is clearly network-based,” he said. He argues that many of the most meaningful signals emerge only when you look at data across relationships: counterparties, business entities, ownership structures, and flows over time. Companies operating alone are always working with an incomplete picture.
However, this change does not remove organizational responsibility. “Companies are still responsible for their customers, their own oversight, and their own decisions,” he said. Therefore, in the future, rather than one model replacing another, both models will merge.
This hybrid model emphasizes data quality. Network intelligence is only as powerful as the underlying inputs. “Even the most sophisticated detections are limited when customer data is weak, entity structures are incomplete, or personal information management is outdated.”
In reality, Nice argues, the effectiveness of network-based detection is rooted in due diligence discipline and continuous monitoring at the enterprise level.
What companies must prove to regulators
For Nice, regulatory trust depends less on the model itself than on the environment surrounding it.
“What companies really need to prove is something very simple but not trivial: that the use of AI is controlled, understandable, and truly improves outcomes.” Claiming efficiency or model sophistication is not enough. Regulators want to see how systems are managed, what data they rely on, how output is reviewed, and how performance is tracked over time.
The real test, he suggests, is whether it can actually be explained. “Can the company explain why the model flagged something, how it was considered, what evidence was considered, and how the final decision was reached?”
Additionally, companies must demonstrate ongoing controls, including keeping data up-to-date, monitoring for drift and bias, and responding to issues as they arise. “Trust comes not from the sophistication of the model itself, but from the control environment surrounding the model.”
Ultimately, Nice believes AI will only be fully accepted once it no longer resembles a black box. Instead, it must function as a “well-managed part of a broader compliance process” that is transparent, accountable and integrated into a company’s decision-making framework.
More needs to be done
Iain Armstrong, executive director of FCC strategy at ComplyAdvantage, said rules-based transaction monitoring is designed to capture known patterns, typologies that the industry has already identified and codified.
He explained, “Machine learning significantly improves this. It detects statistical anomalies rather than named patterns, reducing the amount of false positives and greatly benefiting operations teams.”
But Armstrong said he would like to see more recognition of the fact that money laundering relies on layering and dispersion across institutions and jurisdictions.
“Advanced machine learning models in a given PSP still only know about that PSP’s data,” he said. “We need to make greater use of these technologies in public and private intelligence sharing.”
Regarding the issue of human judgment, Armstrong emphasized that AI shifts the cognitive load from detection to evaluation, which he believes is a real improvement.
“But rather than replacing analysts, it raises the bar for analysts. ‘Reasonable cause to suspect’ is the legal standard, but you still need someone to back it up,” Armstrong concluded.
structural changes
In the view of IMTF co-CEO Sebastian Hetzler, financial crime detection is undergoing a structural shift from static, rules-based surveillance to a more dynamic, intelligence-driven approach.
“At the same time, modern rule-based systems have evolved significantly to achieve high accuracy and low false positives through comprehensive integration of diverse AFC data, so they remain essential to ensuring consistent baseline coverage.”
For Hetzler, machine intelligence complements this by identifying complex patterns and relationships across large datasets, making it particularly effective at detecting both known and new typologies.
He said, “The real value lies in combining both approaches. Rules provide control and protection mechanisms, and AI enhances detection through scale, adaptability, and contextual insights. The most effective approaches combine both, supported by human expertise, and ensure context, interpretation, and accountability.”
Hetzler concluded, “AI is not replacing rules; it is redefining detection. Rules provide a necessary safety net, while machine intelligence brings the scale and adaptability needed to detect increasingly complex financial crimes.”
Get all the latest FinTech news here
Copyright © 2026 Fintech Global
