AML at a regulated bank: 80% of analyst hands-on time automated, with all processing on-premises
AutoCEN by Frontier Foundry — Financial Crime Intelligence for Regulated Banks.
80%
AML analyst hands-on time automated
80%
Reduction in false positives in the first year
100%
On-premises AI inference
<1 day
Median alert-to-disposition time, from more than 2 weeks; many under 1 hour
The Cost of Fighting Financial Crime With Legacy Tools
Our client — a large, regulated banking institution with substantial transaction volumes and a mature AML function — was doing everything right by conventional standards. Experienced investigators. Documented detection rules. Established escalation workflows.
And they were still missing things.
Their compliance analysts were spending the majority of their time on work that should have been automated years ago — triaging alerts from a rigid rules engine, re-verifying entities across disconnected systems, and hand-building SAR narratives case by case. When genuinely suspicious behavior did surface, it competed for attention with a firehose of alerts that were almost entirely benign.
Three challenges were affecting their compliance program:
Analyst overload at scale
Cases and reports were arriving faster than the team could meaningfully review them, forcing triage by volume rather than depth — the conditions under which sophisticated structuring and layering behavior is most likely to slip through.
Legacy tools producing unsustainable noise
The existing rules-based platform was generating false positives at 90%+ rates, meaning the overwhelming majority of analyst hours were spent investigating activity that posed no real risk.
Cloud AI was not an option
The bank evaluated several commercial AI vendors and disqualified every cloud-hosted solution on data-residency grounds. Transmitting customer transaction data to external inference servers was inconsistent with its regulatory obligations and created unacceptable examination exposure.
Challenge vs. AutoCEN Solution
The Frontier Foundry Approach: AutoCEN
Frontier Foundry deployed AutoCEN: a fully on-premises financial crime detection platform engineered from the ground up for institutions that cannot — and should not — move sensitive customer data outside their own infrastructure.
AutoCEN handles detection, investigation and reporting in one system, built for institutions that keep transaction data on their own infrastructure. It is trained on the client's historical transaction data and produces analysis that can be explained and reproduced.
Secure Data Foundation with Kundi
All data ingestion and model training flows through Kundi, Frontier Foundry's sovereign AI platform. Client transaction data is encrypted at rest and in transit, stored exclusively on-premises, and used exclusively to train models within the client's own environment. The system has no connection to external inference services. Data stays inside the client's network by design, and the audit log records every operation.
Rules and Specialized Models, Not One General-Purpose AI
Where commercial AI solutions attempt to apply a single generalist model across every detection task, AutoCEN blends readable rules with versioned models — each model trained on specific AML detection patterns, risk typologies, and behavioral signals. The result is detection that reflects the actual complexity of modern money movement rather than averaging it into a single undifferentiated score.
Platform Capabilities
Real-Time Transaction Surveillance
Screens every transaction in real time against 20+ AML-specific detection patterns — structuring, velocity anomalies, geographic risk, and related typologies — replacing batch-based review with continuous coverage.
Multi-Architecture Detection Ensemble
Readable rules and versioned models analyze behavioral features per event, combining outputs into a single explainable risk score — reducing false positives by 80% in the first year.
Limni Natural-Language Compliance Interface
Through Limni's natural-language interface, analysts ask questions about case data and draft summaries in plain English. The language model handles the question and the wording; detection and scoring come from AutoCEN's deterministic models.
AI-Assisted SAR Narrative Generation
Generates SAR narrative drafts directly from the investigation record, with mandatory analyst review and multi-signature approval required before filing.
Grey-Area Decision Support
For cases below automated-resolution thresholds, compiles supporting evidence and produces a data-backed recommendation that investigators can accept, override, or escalate — every decision preserved in the audit trail.
Versioned Model Updates
Analyst dispositions are recorded as labeled outcomes, and approved model updates ship as versioned releases — without any of that learning being exposed to external systems.
How the models are updated
AutoCEN was deployed on-premises within the client's existing infrastructure, with no disruption to the compliance team's workflow. Within the first 12 months:
80%
of AML analyst hands-on time automated — analysts redirected from repetitive review to judgment-intensive casework
80%
reduction in false positive alerts — each hour of analyst time now yields substantially more compliance value
↓
reduction in AML operating cost in the first year
100%
data-backed decision support on grey-area cases — compiled evidence and explainable recommendations for every human review
Analyst dispositions are recorded as labeled outcomes. The bank's model-risk team reviews proposed model updates, and approved updates ship as new, versioned releases that can be replayed against earlier cases. SR 26-2, which superseded SR 21-8 in April 2026, is now the model-risk guidance that applies to BSA/AML systems; see our SR 26-2 validation pack.
Why Regulated Banks Choose Frontier Foundry
Frontier Foundry's founder served as the FDIC's inaugural Chief Innovation Officer, reporting directly to the Chairman.
AutoCEN is currently deployed across regulated financial institutions, including top-20 U.S. banks. Results describe one client's experience over the period stated and are not a guarantee of future results. Past performance is not indicative of future results. Client identities are withheld under NDA; references are available under NDA. Frontier Foundry has completed a SOC 2 Type II examination covering a 2026 examination period. The report and auditor details are available under NDA.
Ready to See AutoCEN in Your Compliance Function?
Request a briefing. We'll walk through how this deployment worked and what it would take in your environment.