Pilot Specification #1 — BSA/AML Sanctions Alert Triage
Heritage Community Credit Union
Journey deliverable · Seiche Advisors
Version: 1.0 · August 2026
Status: Board-authorized pending POC contract
1. Pilot summary
| Field | Value |
|---|---|
| Use case | AI-assisted triage of OFAC/sanctions alerts before analyst review |
| Primary vendor (recommended) | Nasdaq Verafin — agentic sanctions review module |
| Duration | 90 days (POC) + 30-day evaluation |
| Start target | November 1, 2026 |
| Executive sponsor | Diane Foster, BSA Officer |
| Technical owner | James Wu, CIO |
| Budget (POC) | $45K–$65K (license + services) |
2. Problem statement
HCCU BSA team processes ~1,400 alerts/month across sanctions screening and transaction monitoring. Internal time study (June 2026) shows:
- 62% of analyst time on false-positive disposition
- 18% on true-positive investigation
- 20% on SAR drafting and QA
Sanctions alerts average 47 minutes each to clear; 73% are false positives. At 4 FTE analysts, backlog risk is increasing as wire volume grows.
3. Pilot objective
Demonstrate that AI-assisted sanctions alert triage can reduce analyst time per alert by ≥40% without increasing missed true positives, measured against a 90-day baseline.
4. Scope
In scope
- OFAC/sanctions alerts generated by existing Abrigo BAM+ workflow
- Alert enrichment, duplicate detection, and recommended disposition (escalate / clear)
- Analyst review and override of all AI recommendations (human-in-the-loop mandatory)
- MRM documentation: model card, validation summary, override log
Out of scope
- Full AML transaction monitoring replacement
- Autonomous SAR filing
- Member-facing communications
- Credit decisioning
5. Success metrics
| Metric | Baseline | Target (90-day) |
|---|---|---|
| Avg. minutes per sanctions alert | 47 min | ≤28 min |
| False positive rate (unchanged or lower) | 73% | ≤73% |
| Missed true positives | 0 in baseline quarter | 0 |
| Analyst overtime hours/month | 38 hrs | ≤20 hrs |
| Override rate (AI vs. analyst) | N/A | Track; expect 15–25% initially |
6. Data & integration
| System | Role |
|---|---|
| Symitar | Source of member and transaction data |
| Abrigo BAM+ | Alert origination; disposition workflow |
| Verafin | AI triage layer (API integration) |
Data classification: Restricted — member PII, transaction data. Processing in vendor SOC 2 environment; no training on HCCU data per contract Exhibit B.
7. Governance & compliance
- Human-in-the-loop: No alert closed without analyst sign-off in POC phase
- Model risk: Documented in AI inventory; annual validation per SR 11-7 principles
- Fair lending: N/A for BSA use case
- Examiner narrative: One-page POC summary for NCUA annual exam file
8. Timeline
| Week | Activity |
|---|---|
| 1–2 | Contract, DPA, vendor security review |
| 3–4 | Integration build, test environment |
| 5–6 | Parallel run (AI recommends, analysts decide without AI influence) — baseline |
| 7–14 | AI-assisted triage live; daily override review |
| 15–16 | Results analysis, board memo, scale/stop recommendation |
9. Budget
| Line item | Amount |
|---|---|
| Verafin POC license (90 days) | $28K–$38K |
| Implementation / integration | $12K–$18K |
| Seiche PMO (fractional) | $5K |
| Total POC | $45K–$61K |
Scale decision (if successful): Full annual license $95K–$130K — included in Year 2 budget request.
10. Risks & mitigations
| Risk | Mitigation |
|---|---|
| Integration delay with Symitar | Jack Henry vendor management call Week 1 |
| Analyst resistance | Co-design workflow; Foster leads change mgmt |
| Vendor model update | Contract notification clause; re-validation trigger |
11. Scale / stop criteria
Scale if: ≥40% time reduction, zero missed TPs, override rate stabilizes below 20% by Week 14.
Stop if: any missed true positive, examiner objection, or <20% time reduction with no improvement trend.
Approvals
| Role | Signature | Date |
|---|---|---|
| BSA Officer | ||
| CRO | ||
| CIO | ||
| CEO |
© 2026 Seiche Advisors, LLC. All rights reserved. · seicheadvisors.com · Proprietary & Confidential