Anti-money laundering (AML) investigations are among the most high-stakes, high-volume processes in financial services — and they’re notoriously slow, manual, and fragmented. Investigators are stretched thin trying to reconcile Know Your Customer (KYC) records, transaction alerts, graph connections, and external watchlists, all while under regulatory pressure to move fast and document every step.
In our most recent webinar, John McCambridge, Global Director of Financial Services Solutions at Dataiku, and David Behar, Senior Manager of Data Science Business Solutions, revealed how AI agents built in Dataiku can eliminate these bottlenecks.
The star of the session? The AML investigator assistant, an agent-powered architecture in Dataiku that automates risk identification, flags anomalies, prepares audit-ready case summaries, and even drafts Suspicious Activity Report (SAR) elements — without compromising control or compliance.
In this recap, we’ll walk through what they shared, spotlight the full demo, and show why this is one of the most compelling use cases for AI agents in financial services today.
What's the Big Deal With AI Agents?
John opened the session with a powerful framing: AI agents are not rule-based bots. They’re decision support systems that analyze signals across silos, generate insights, and collaborate with human experts. Especially in AML, where alert volumes are rising and compliance costs are climbing, agents offer a scalable path forward.
Here’s what AI agents with Dataiku deliver for AML and Countering the Financing of Terrorism (CFT) teams:
- Real-time risk aggregation from transactions, KYC data, and graph intelligence
- Context-aware decision recommendations based on alert types and entity behaviors
- Auto-generated case packets that reduce SAR prep time from hours to minutes
- Fully governed workflows that auditors and compliance teams can trust
Inside the Demo: Building the AML Investigator Assistant
David’s demo was the heart of the webinar — a step-by-step tour of how an AML agent system is built and deployed. Using visual tools inside Dataiku, David showcased how multiple agents work together in a unified investigation flow.
Building Blocks: Specialized Agents
The webinar zoomed out to showcase the entire agentic architecture behind the AML investigator assistant. Here are the agentic components that power it:
All of these agents are orchestrated inside Dataiku — connected via tools like:
- Dataset Lookup Tool (for account info, alerts, and statements)
- Web Search Tool (to pull external context like news sentiment)
- Graph Exploration Tool (to create interactive visualizations)
The Flow: From Alert to Action
We didn’t just build an agent that searches. We built an assistant that thinks through the case and hands the investigator everything they need.
-David Behar, Senior Manager of Data Science Business Solutions at Dataiku
The system begins with a flagged transaction. From there:
- The AML assistant agent assembles the full context — KYC history, transaction behavior, known associations, and external risk signals.
- It generates a structured risk summary report, delivered via email and dashboard, calling out high-priority factors and surfacing hidden threats.
- It engages the AML investigator through a chat interface, enabling a deeper dive into suspicious patterns.
- If the investigator believes that the pattern is truly suspicious, the SAR agent kicks in, pulling from templates and David’s custom logic to draft report-ready elements.
This isn’t just fast. It’s complete, explainable, and compliant.
Why This Changes the Game for AML-CFT Teams
AML-CFT teams are drowning in disconnected systems, skyrocketing alert volumes, and relentless SAR deadlines. Investigators waste hours stitching together KYC data, transaction history, and risk signals just to get a basic read on a case.
The AML investigator assistant changes that. It pulls together fragmented data, pinpoints key risk factors, and auto-generates structured, audit-ready reports. Investigators get a full picture upfront with no context hunting or manual rework necessary. And because agents are explainable and fully governed, teams stay in control while moving faster and smarter.
Here's What You Should Take Away
1. AI agents = actionable intelligence: AI agents surface insights, not noise, and help investigators act faster.
2. Human remains in control: Agents accelerate decision-making while investigators maintain full ownership over analysis, conclusions, and final sign-off.
3. Built entirely in Dataiku: Every part of the architecture — from data ingestion to SAR prep — lives in one governed, composable platform.
4. Designed for reuse: This isn’t a one-off project. The agents and architecture here can be reused across:
- Sanctions screening
- Client coverage
- ESG scoring
- Fraud detection
This is the type of intelligent automation financial institutions have been waiting for.
-John McCambridge, Global Director of Financial Services Solutions at Dataiku