RL Environment Reference: Financial Services
A complete workflow map of how private banks (JPMS, Goldman Sachs Private Wealth, Morgan Stanley) onboard ultra-high-net-worth clients. 11 phases, 15+ systems, 6 regulatory frameworks. This is the kind of enterprise workflow that makes a high-value RL training environment.
Architecture
Wealth onboarding is not one system. It's 10-15 platforms orchestrated together. The onboarding orchestrator is the brain — it coordinates all systems end-to-end.
Workflow Phases
Click any phase to expand the detailed workflow, system interactions, regulatory requirements, and decision gates.
CIP = Customer Identification Program, mandated under the USA PATRIOT Act (Section 326) and equivalent regulations globally. The bank must form a "reasonable belief" of the client's true identity.
KYC builds a comprehensive risk profile of the client — who they are, where their money comes from, what they want to do with it, and what patterns of activity to expect.
Every client (and every UBO, related party, and authorized signatory) must be screened against global watchlists before any account is opened. This is non-negotiable under AML regulations worldwide.
No account opens without a chain of internal approvals. The approval chain varies by client risk tier and product complexity.
AML compliance is NOT a one-time event. It is a continuous obligation for the lifetime of the relationship. Failure to monitor results in regulatory penalties, fines, and criminal liability.
Regulatory Landscape
| Framework | Scope | Key Requirements |
|---|---|---|
| BSA / AML | US anti-money laundering | CIP, CDD, SAR/CTR filing, ongoing monitoring, recordkeeping |
| FATF Recommendations | Global AML/CFT standards | Risk-based approach, CDD, EDD for PEPs, wire transfer rules |
| USA PATRIOT Act | US counter-terrorism | Section 326 (CIP), Section 312 (EDD for foreign banks), Section 314 (information sharing) |
| OFAC Sanctions | US sanctions enforcement | Screen all parties against SDN list; block/reject prohibited transactions |
| FATCA | US tax compliance (global) | Identify US persons; report account balances to IRS; W-8/W-9 collection |
| CRS (OECD) | Global tax transparency | Self-certification; automatic exchange of financial account info between 100+ jurisdictions |
| EU 6AMLD | EU anti-money laundering | CDD/EDD, beneficial ownership registers, criminal liability for AML failures |
| MiFID II | EU investment services | Suitability assessment, best execution, product governance, cost transparency |
| Reg BI (SEC) | US broker-dealer standard | Best interest obligation, disclosure, care, conflict of interest mitigation |
| GDPR | EU data privacy | Lawful basis for processing, data minimization, right to erasure, breach notification |
| AMLA 2020 | US AML modernization | Beneficial ownership reporting to FinCEN, whistleblower protections |
RL Environment Potential
This onboarding process is exactly the kind of enterprise workflow that current AI models struggle with — and that makes it valuable as an RL training environment.
| Dimension | What Makes It Hard |
|---|---|
| Multi-system orchestration | 15+ systems that must coordinate sequentially and in parallel; failures cascade |
| Regulatory complexity | 11 overlapping regulatory frameworks across multiple jurisdictions with conflicting requirements |
| Decision gates | Binary pass/fail screening + multi-tier risk classification + multi-stakeholder approval chains |
| Data complexity | Beneficial ownership tracing through layered corporate structures across jurisdictions |
| Exception handling | PEP escalations, EDD deep dives, conditional approvals — the happy path is only 40% of cases |
| Human judgment | Fuzzy name matching disposition, risk classification overrides, "reasonable belief" standards |
| Time pressure | Clients expect 2-week onboarding; compliance requires thoroughness; RM wants speed |
| Audit requirements | Every decision must be documented with rationale; full audit trail for regulatory examination |