We built an AML platform that uses graph neural networks to trace multi-border transaction networks, filters out benign activity and drafts evidence briefs for investigators.
Industry: FinTech & Financial ServicesService: AI & Automation
Overview
Client
HSBC
Global banking and financial services group
Engagement snapshot
The challenge
Rules-based systems looked at transactions one at a time, so they missed criminal networks spread across many hops and borders.
At the same time, they raised large numbers of false-positive alerts, and investigators spent much of their day clearing benign activity.
Multi-hop money movements slipped past single-transaction rules.
False positives buried the cases that mattered.
Evidence for regulatory reports was gathered by hand.
The solution
Graph neural networks analyse complex, multi-border transaction pathways in real time.
Normal money movements are filtered out, and high-probability illicit networks are prioritised.
Agentic AI drafts structured regulatory reports with transaction evidence for analysts to review.
Key features
Graph neural networks follow money across complex, multi-border transaction chains in real time.
Business impact
Hidden networks uncovered
Benign money movements are filtered out before alerts reach the team.
Business impact
40% fewer false-positive alerts
Agentic AI drafts structured compliance reports with transaction evidence for review.
Business impact
Faster investigative review
Before and after
Before
One transaction at a time
Rules missed networks spread across many hops.
After
Whole networks
Graph models trace money across hops and borders.
Before
Mostly noise
Benign activity triggered large numbers of alerts.
After
40% fewer false positives
Benign movements are filtered out first.
Before
Assembled by hand
Analysts gathered evidence manually.
After
Drafted with evidence
AI drafts structured reports for analysts to check.
Results
40% Fewer False Positives
False-positive AML alerts dropped by 40%.
Before: Rules OnlyAfter: Graph AI
Hidden Networks Found
Graph analysis uncovers hidden transaction networks that rules-based systems missed.
Before: Single TransactionsAfter: Whole Networks
Faster Review
Drafted evidence briefs accelerate the investigative review process.
Before: Manual EvidenceAfter: Drafted Briefs
Testimonial
“The graph-based AI engine uncovers hidden transaction networks that traditional rules-based systems missed. We've reduced false-positive AML alerts by 40% while accelerating our investigative review process.”
Richard BlackburnGroup Chief Risk and Compliance Officer, HSBCKey results
Trust and control
Investigators review every network and approve every report.
HUMAN IN CONTROLReports show the transaction trail behind each finding.
EXPLAINABLEAlerts, decisions and reports are logged for regulators.
TRACEABLEModel versions and changes are documented.
GOVERNEDFaster delivery
Reusing tested components let the team focus on the bank's own typologies and reporting formats.
Tested components turned transaction streams into graph data quickly.
Versioned models and documented features support compliance review.
A ready-made queue routes networks to investigators.
Every alert, decision and report is recorded.
We built an automated anti-money laundering and compliance workflow. Graph neural networks analyse complex, multi-border transaction pathways in real time, benign money movements are filtered out, and high-probability illicit networks go to human investigators with evidence briefs drafted by agentic AI for them to review and approve.
Rules-based systems judge transactions one at a time, so normal activity often looks suspicious. Graph neural networks look at how money moves across a whole network of accounts and hops, which separates benign movements from illicit patterns. False-positive AML alerts dropped by 40% while investigative review became faster.
No. Agentic AI drafts structured regulatory compliance reports complete with the transaction evidence, but analysts review and approve every report before anything is filed. Investigators also decide on every network the platform surfaces, so the bank keeps full control over its compliance decisions and regulatory filings.
If your AML team spends its day clearing false positives, we can build graph AI that surfaces real risk.