CASE STUDY AML • Name Screening

Name Screening Engine Tuning

Phase II tuning of the AML name screening engine to reduce false hits and improve operational throughput for Retail & Corporate Banking under tight regulatory timelines.

ALM • Actimize Name matching / Fuzzy Matching Agile Delivery L3 Production Support Regulatory Compliance

30-second skim

Fast highlights / Quick scan

Role

Product Enablement Lead (Scrum Master + BA)

Tech

NICE Actimize, Jira, Confluence

Goal

Reduce the Name-Screening engine's false positive rate (FPR) to enhance manual review throughput without increasing compliance exposure.

Challenge

Inconsistent data baselines + complex regional naming logic + hit-rate sensitivity tuning + high L3 production support load

Outcome

25–30% ↓ Rework • Reproducible tuning baseline (runbook + test set) • Stabilized velocity • Maintained momentum during BAU noise

Artifacts

Decision log • Forensic Triage Lane • Risk Register • Refined Backlog

Problem & why it mattered

High false-positive hits created operational drag for stakeholders and increased regulatory risk.

The "Silent Failure" of Data Drift: Inconsistent engine states and "Point-in-Time" data shifts made tuning results non-reproducible. Calibrating logic for regional naming variations was impossible without a stable "frozen" baseline.
The Operational Risk: Without a stable laboratory environment, we risked making tuning decisions on shifting data, potentially leading to missed sanctions or total manual review overload.
Resource Contention: SME (Subject Matter Expert) velocity was pulled into emergency L3 production support and "Orphaned Assets" (legacy apps from other squads). Parallel high-priority migrations and EOL upgrades added further cognitive load.

Workflow (before → after)

Transitioning from reactive "firefighting" to a forensic, data-first delivery model.

Fragmented Operational Workflows (Before)

Optimized Product Infrastructure (After)

Tuning based on contaminated data snapshots → high rework in end-cycle testing.

Boolean Gap Analysis + Zero-Test ensuring a stable, 100% reproducible baseline.

SMEs hijacked by invisible L3 support fires; velocity stalled.

Explicit Forensic Triage Lane in Kanban made support visible.

High noise due to misaligned regional naming parameters.

Aligned regional naming logic with engine sensitivity parameters.

Reactive culture; 30%+ bandwidth lost to unmanaged "Dark Matter."

Refined backlog + shielded velocity reduced delivery rework by 25-30%.

Approach & Decisions

Structuring a data-driven framework to transform a reactive environment into a high-precision delivery pipeline.

What I did (key moves)

Strategic Discovery: Led a "Double Diamond" process to map complex engine logic and design a forensic tuning framework from the ground up.
The Data Pivot: Identified "Silent Data Drift"; halted tuning to resolve inconsistencies via Boolean Gap Analysis, ensuring a stable, 100% reproducible baseline.
Operational Shielding: Protected squad velocity by institutionalizing a Forensic Triage lane to neutralize support "hijacks" and protect core development.
Regulatory Logic: Aligned Entity-Centric scoring (RP Linking) and cultural profiles while maintaining an audit-ready Decision Log for authorities.

Decisions / trade-offs

Data integrity first: Tuning decisions only proceeded after reproducible baseline controls were in place.
Operational visibility over heroics: Made hidden support load explicit and triaged to protect planned delivery scope.
Auditability by design: Maintained decision logs and risk tracking so optimization trade-offs remained explainable under compliance scrutiny.

Delivery under Constraints

High-velocity execution despite hyper-utilization and external dependencies.

Constraints (top 3)

Forensic Triage: Made untracked BAU/support work visible (triage lane + prioritization rules).
Asset Repatriation: Offloaded legacy debt.
Workstream Orchestration: Balanced hyper-utilization.
Quality Guardrails: Minimized context-switching.
Active Risk Register: Neutralized silent blockers.

Name screening delivery constraints and execution

Impact

Strengthened regulatory confidence and stabilized the delivery trajectory for a core AML system, reducing rework by 25-30% through shared visibility and forensic process optimization.

Rework Reduction

25-30% decrease in delivery rework through strategic backlog management and "Zero-Test" validation.

System Precision

Reproducible baseline across historical runs (runbook + fixed test set).

Operational Governance

Aligned cross-functional sub-teams and synchronized downstream dependencies for stable release flow.

Agile Leadership & System Optimization

Stabilizing the development lifecycle to enhance the performance of a core fintech engine.

Algorithmic Calibration: Improved alert quality (signal-to-noise) by reducing false positives in priority scenarios.
Strategic Prioritization: Stabilized development flows.
Operational Governance: Aligned cross-functional sub-teams.
System Orchestration: Synchronized downstream dependencies.

Next improvements (post-launch roadmap)

Observability: Logs existed, but insight was hard to surface. Make the system operationally observable so evaluation can track inputs, outputs, and drift as quality evolves.
Governance: Continue strengthening asset ownership and exit strategies to keep delivery capacity protected.

Learnings

Strategic takeaways from fintech logic and product operations.

Operating lessons

Data Integrity: Tuning is a guessing game without a forensic data baseline.
Visibility: "Dark Matter" kills velocity; shared visibility aligns bandwidth.
Trade-offs: Optimizing a probabilistic engine means balancing false positives vs false negatives with audit-ready decisions.
Governance: Scale requires rigid asset ownership and clear exit strategies.
Observability: Logs existed, but insight was hard to surface. Make the system operationally observable so evaluation can track inputs, outputs, and drift as quality evolves.
Trajectory. Success means architecting the framework and stabilizing the path.