Skip to main content
Case Study · Compliance & RegTech

AI Compliance Scoring Across 180+ Jurisdictions

Building real-time, explainable AI risk scoring for cross-border transactions — where AI advises but never silently blocks.

Results

Real-time
Screening
-60%
False positives
100%
Decision explainability
180+
Jurisdictions

Technology Stack

  • Rules engine
  • AI detectors
  • Sanctions datasets
  • Append-only audit log
  • Event-driven triggers

The Challenge

The platform screened counterparties and transactions manually against sanctions and watch lists. Updates were missed, false positives buried analysts, and — most dangerously — the team could not explain to a regulator *why* a given transaction had been cleared or held. In a regulated business, an unexplainable decision is a liability.

The Solution

We built a hybrid engine that combines deterministic rules with pluggable AI detectors. Rules handle the unambiguous cases — a sanctioned party is always blocked. AI detectors add nuance: jurisdiction risk, route anomalies, valuation mismatches, and document gaps, each producing a confidence score. A fusion layer corroborates signals across both, and crucially, AI alone can never silently block a transaction — it raises it for human review. This is the production philosophy described in cross-border compliance AI.

Architecture

A rule analyzer evaluates each transaction against canonical, versioned lists. Pluggable AI detectors run in parallel, each emitting findings with confidence. The fusion stage combines them, applying corroboration boosts and data-gap discounts. Every step is written to an append-only audit log so the full reasoning chain is reconstructable. Screening is triggered by platform events, so it happens in real time as transactions flow.

Technology Stack

A deterministic rules engine, a registry of pluggable AI detectors, versioned sanctions datasets, an append-only audit store, and event-driven triggers. The work draws on our AI solutions and AI agents practices and underpins the real-time settlement path.

Results

Screening moved from batch to real time, false positives fell by roughly 60%, and every decision now carries a complete, exportable reasoning chain. Coverage spans more than 180 jurisdictions, with new rules and detectors added without redeploying the core.

Lessons Learned

The single most important design rule was that AI advises but never silently blocks — it escalates to a human. Pairing deterministic rules with probabilistic detectors gave both certainty where it mattered and nuance where it helped. And an append-only audit log turned compliance from a defensive scramble into a routine export.

Frequently Asked Questions

Can the AI block a transaction on its own?+

No. Deterministic rules block unambiguous cases; AI detectors raise risk for human review but never silently block. This keeps decisions defensible.

How are decisions made auditable?+

Every rule evaluation and AI finding is written to an append-only log, so the complete reasoning chain behind any decision can be reconstructed and exported.

How are new sanctions lists handled?+

Lists are versioned datasets; updates flow in without code changes, and re-screening can be triggered for affected transactions.

What makes the hybrid approach better?+

Rules give certainty for clear cases; AI adds nuance for ambiguous ones. Fusing both, with corroboration, reduces false positives without missing real risk.

More case studies

Achieve outcomes like these

Talk to our strategy team about how the Baalvion Operating System can power your next platform.