XYZ BANK PUBLIC COMPANION

AI-POWERED AML NAME SCREENING

Remove the noise.
Keep the signal.

Explore how a controlled screening workflow can reduce avoidable manual review without trading away known true-hit retention.

PLAY — MATCH THE NAMES

Should this alert be investigated?

Choose the action you think is safe. These are simulated name pairs for demonstration only.

WATCHLIST PARTY CHAN TAI MAN 陳大文
SIMULATED CUSTOMER CHAN TAI MAN 陳大文

SIMULATED CUSTOMER — DEMONSTRATION ONLY

Select an action to inspect the control decision.

PLAY — WHY SMALL IMPROVEMENTS MATTER

How much difference can two points make?

ILLUSTRATIVE EXAMPLE — NOT ACTUAL BANK DATA

Relative review workload25.0
Change from 96%—

Assumes unchanged true-hit volume. At 94%, relative workload is 16.7 — about 33.3% lower than at 96%.

UNDERSTAND — HOW THE SYSTEM WORKS

A controlled analytical layer.

  1. 01Name Pair
  2. 02Normalize
  3. 03Semantic Representation
  4. 04Similarity Score
  5. 05Controlled Decision Boundary

Human investigators retain ambiguity, identity verification, context, and final AML responsibility.

RESULTS — 100%: WHAT IT MEANS

A control requirement,
not a claim of perfect AI.

IT MEANS

Every known true hit in validation must remain in the investigation flow.

IT DOES NOT MEAN

AI can never miss any future risk.

EVIDENCE

Explore the control framework.

Evidence is organised around business controls, not marketing metrics.

ValidationThreshold ControlHuman OversightAudit TrailOngoing MonitoringGovernance