AI Agents

Context

Built for banking within the Financial crime risk assessment stack, this agent is applied at onboarding and periodic refresh—and continuously on higher-risk books—to detect accounts that shift from low/no activity to sudden high velocity inconsistent with the profile on file

What it does

The agent ingests recent account history and constructs a baseline for balances, credits/debits, counterparties, and channel/geo use. It then monitors for reactivation patterns: sharp step-ups in volume immediately after dormancy, bursts of inbound credits followed by rapid dispersals, new or concentrated counterparties, changes in channel mix (e.g., ATM/cash, cross-border wires), and geo anomalies relative to the customer footprint. For each spike, it returns a clear finding with a short rationale, confidence, and record-level links to the specific transactions and windows that triggered the alert, so investigators can see what changed, when, and why it matters.

Core AI functions

Baseline construction over multi-period histories; change-point and burst detection on balances and flows; cadence analysis to distinguish genuine re-engagement from mule-like pass-through; counterparty novelty and concentration scoring; channel and geography deviation checks; peer/profile coherence testing; and reason-code generation with line-of-sight to the exact ledger rows, counterparties, and timestamps cited.

Problem solved

Dormant accounts that suddenly reactivate are a known conduit for fraud and AML risk, but manual reviews catch them late or generate noise from benign restarts. Evidence is dispersed across systems, slowing investigations and producing uneven narratives.

Business impact

Earlier interdiction with fewer false positives. Material reactivation spikes surface within the review window, escalation includes click-through evidence for workpapers and SAR drafting, and benign restarts are filtered by profile/peer context. Outcomes are more consistent, audit trails are cleaner, and supervisory confidence improves.

Integration and adjacent use cases

Integration complexity: Low to medium. The agent reads account activity from your core/ledger or AML data store and writes findings, rationales, and evidence links back to onboarding/refresh or investigation workflows; core systems remain unchanged.

It is commonly combined within this stack with

• Behavioral risk scoring agent to convert findings into portfolio-level tiers,

• Cash transaction pattern detector to separate cash-driven bursts from normal restarts,

• High-risk merchant category detector where category drift may explain or contradict the spike, and

• Anomalous peer network growth to reveal emerging rings around the reactivated account.