Today, I’m excited to share that Hummingbird has acquired Quantimon to build transaction monitoring for the agentic era. Founded by industry veterans from Apple, PayPal, and Xoom, Quantimon sharpens AML risk detection with machine learning models that learn from alert outcomes and through a federated learning network. Paired with Hummingbird’s investigations platform, our combined solutions will deliver what AML has long been missing: detection that improves with every decision a compliance team makes.
Most AML transaction monitoring still runs on rules: they fire, the alerts pile up, and most turn out to be nothing. Machine learning promised a fix that has gone largely unrealized. That’s because models need labeled outcomes to learn from – whether the activity being alerted on was suspicious and why – and until now, those outcomes have lived in case notes, scattered across systems and too expensive to turn into training data.
AI changes both sides of this problem:
AI agents can work far more alerts than people alone can, while ensuring every finding is explainable and documented. Teams can run the broad rules that a strong program needs without building a backlog nobody can clear.
Machine learning risk-ranks those alerts so teams focus on behavior that’s most likely to be suspicious. And because AI can turn every resolved alert into well-labeled training data, each decision feeds back into detection.
This loop only works when detection and investigation run on the same platform. Quantimon built models designed to learn from alert decisions, and Hummingbird is where those decisions get made every day. Together, models surface the risk, AI agents and analysts collaborate to resolve it, and every resolution improves the next round of detection.
The Quantimon team is joining Hummingbird to continue their work, now as part of enhancing Hummingbird’s monitoring solution. As the team gets started, customers can expect:
Richer rule-based detection: Making rules easier to build, test, and manage, with support for larger and more complex monitoring programs. Machine learning works best layered on strong rule coverage, not in place of it.
Machine learning models trained on your alert outcomes: Models that learn from what your team decides to close, escalate, or file on, enabling you to rank alerts by risk and catch activity that resembles your confirmed cases even when no rule fires.
A network that sees beyond your own walls: Over time, you will be able to draw on signal from outside your institution – from third-party intelligence or patterns across the wider Hummingbird network – so you can catch risk you didn’t know to look for.
Hummingbird customers already run transaction monitoring and investigations in one place, with AI agents and analysts working alerts side by side. What’s been missing is a way for their decisions to actually sharpen the next round of detection. Bringing Quantimon into Hummingbird closes that gap. For compliance teams, this means less time clearing false positives and more time investigating activity that warrants real attention.
If you're excited by today's news or rethinking how your monitoring program should work in an agentic world, please get in touch. We'd be glad to show you what we're building.
Subscribe to receive new content from Hummingbird
