Beyond Hype: The Practical Future of AI in Financial Crime

An Interview with Daoud Abdelhadi, Data Science Lead at Eastnets

AI has been the "hot topic" in finance for years, but recent shifts - from regulatory clarity to the consumer adoption of tools like ChatGPT - have finally pushed financial institutions past curiosity and into serious adoption.

We spoke with Daoud Abdelhadi, Data Science Lead at Eastnets, about the urgent need for adaptive technology, the myth of the "black box," and the powerful role Agentic AI will play in transforming financial crime fighting.

Key takeaways:

• AI adoption is a necessity, not a novelty: The shift is driven by regulatory clarity and the unsustainable rise in false positives.
• The future requires 360-degree risk monitoring: Financial crime systems are traditionally siloed. We need to combine all dimensions of risk into a single holistic view to make better judgments.
• Agentic AI will transform investigations: Beyond traditional machine learning (ML), leveraging reasoning abilities will automate tasks effectively.
• AI augments human teams and can be deployed gradually.
• Explainability and rules co-exist: The solution is not to eradicate rules but to use them for clear cases, while leveraging AI for dynamic behavior. 

The Urgent Shift: Why Banks Are Embracing AI Now

“Automation is no longer a luxury, it’s a necessity”.

The past 12–18 months have marked a critical turning point for AI adoption in compliance and financial crime. For Daoud, this  shift has been driven by three things:

1. Regulatory Green Light:The introduction of frameworks like the EU AI Act provides structure and guidelines, giving banks the confidence to adopt adaptive techniques. 
2. Automation as a Necessity: The sheer volume of global transactions continues to rise. Relying on less sophisticated, static techniques leads to unsustainably high alert rates and false positives. Automation is essential for operational survival.
3. The ChatGPT Effect: In Daoud’s opinion, the accessibility and reasoning power of consumer-facing AI tools have subconsciously educated the market. People are seeing the tangible value of inherent intelligence and are pushing to incorporate it into professional workflows.

The Problem with Static Rules

Financial crime is fundamentally about understanding behavior - how an entity (person or company) transacts within a banking system. Behavior is dynamic: a student behaves differently than a CEO, and a small company operates differently than a large one.

“If we use static rules to monitor behaviour, you’re essentially casting a very wide net on a very dynamic environment. This is a recipe for very high false positive rates. What you need is a system that can adapt to data. The best tool we have to do that is machine learning and AI.”

The Future of AI: Holistic Monitoring and Agentic Intelligence

According to Daoud, there are two areas driving the next evolution of AI in financial crime:

1. Holistic AI Monitoring (The 360-View)

Currently, banks tend to treat risk in silos; AML, fraud, and sanctions screening all operate independently. These, however, are simply different dimensions of risk for the same entity.

The future involves combining all these data points into a single, 360-degree view of risk. For example, knowing that a precursor to money laundering is often fraud allows the AI to make a far better, contextualized judgment. Combining fraud, AML, and sanctions information allows investigators to transition from isolated alerts to understanding the full risk profile of a customer.

2. The Power of Agentic AI

“People use ChatGPT everyday. What they don’t realise is that the key power of it is its reasoning abilities. If you stop thinking about it as a content generation tool and start leveraging the inherent intelligence, you can use it for powerful purposes.”

Agentic AI is revolutionizing investigation workflows.

By training a model to behave like a financial crime investigator, you can automate tasks far beyond what traditional ML or rules can achieve. Agentic AI can:

• Automatically retrieve information from various internal systems and databases.
• Write preliminary drafts for Suspicious Activity Reports (SARs).
• Handle complex triage and investigation tasks that were previously manual burdens.

This is where the combination of rules, traditional machine learning, and Agentic AI takes risk management to a "whole new level."

How Eastnets Deploys AI: Augmentation, Not Eradication

“I think there’s this misconception that now there’s AI, we need to completely eradicate what’s been working moderately well, which is rules. But that’s not the case. Rules are very effective if you have clear cases or scenarios you want to monitor. Where you want to use AI is when you’re dealing with something dynamic”

Where Eastnets uses AI is in tackling two of the biggest pain points reported by banks: false positive reduction and improving the quality of alerts.

1. AI for Optimization (False Positive Reduction)

When a rules-based system generates inaccurate alerts, AI acts as a layer of intelligence, or a second opinion, to minimize those false positives. This uses ML to clean up and optimize the noisy output of existing rules.

2. AI for Detection (Generating Higher Quality Alerts)

Here, Eastnets uses supervised and unsupervised models to alert when something abnormal has occurred:

• Supervised Learning: The AI is taught using examples of known crime/no-crime scenarios. The challenge here is the relatively small size of data sets to learn from. 
• Unsupervised Learning: The AI is allowed to find abnormal patterns within data without predefined examples.

By combining these frameworks, Eastnets builds a robust, multi-layered system capable of catching suspicious behavior.

Navigating the Switch: From Advisor to Automation

“You don’t have to do things all at once. You can take things step by step.”

Adopting AI doesn’t require a wholesale change all at once. Eastnets champions a step-by-step, trust-building approach:

1. Pre-Deployment Analysis (Proof of Concept): We gather the bank’s data and simulate the AI model to provide an immediate glimpse of its performance on real data. Banks can see the value and know exactly what to expect. 
2. Deploy as Advisor First: The AI can be deployed to recommend decisions, but humans remain the final decision-makers. The investigator sees the AI's suggestion and uses it as a supportive tool to make a more accurate decision.
3. Gradual Automation: As confidence builds and the AI proves itself reliable (its decisions consistently mirror human intuition), the bank can gradually increase the level of automated tasks.

The End of the 'Black Box'

“The biggest misconception I see is that AI is touted as this black box model. Whereas over the past few years, there's been a lot of development in research in a practical sense to make AI more explainable.”

At Eastnets, we ensure that the factors behind every AI decision are revealed and supplemented with analytics. 

By showing investigators the historical data the AI is feeding off of, they gain a complete, explainable view of the AI's behavior. As Daoud concludes, "AI won’t replace humans, humans that use AI will replace humans." It’s about leveraging the best tools to perform the task more effectively.

Learn more about SafeWatch Screening, our all-in-one interface to help financial institutions detect and prevent financial crime with AI-driven analytics and false positive reduction.

Sources and references

1. How AI is transforming financial crime detection
2. Global financial institutions turn to AI and automation as trade-based financial crime hits $1.6 trillion annually
3. Interview with Daoud Abdelhadi, Data Science Lead at Eastnets