Project overview
Anti-money laundering (AML) work involves reviewing activity that may deserve a closer look. An alert signals a possible issue; it does not establish wrongdoing. This concept explores how an AI agent could gather context and prepare findings before a bank employee reviews the evidence and decides what to do.
The opportunity
The proposed direction extends AI assistance beyond a summary: an agent could work through supporting information and prepare a review. The design challenge is to make that work visible and understandable, so a reviewer can distinguish unfinished research from a completed recommendation.
Show what the agent is doing
The alert list introduces visible states: Pending, Researching, Complete, and Not Applicable. Completed rows show a recommendation such as Clear, Watch, or Send to Case. The status and recommendation occupy separate columns, distinguishing progress from the suggested next step.
Bring the reasoning into the review
The detail screen combines a recommendation, a summary, and supporting evidence. Additional panels organize scenario insights, customer context, previous activity, and account information. An Agent Findings card includes a View Evidence action, giving the reviewer a path from a concise finding to its supporting material.
Help with documentation
The comments area includes an Agent [Draft] label above a proposed explanation. This presents the text as a starting point for review. The concept also retains the main Clear and Send to Case actions, rather than hiding the decision behind the agent’s summary.
Questions to validate next
The next step would be to test whether reviewers understand the research states, distinguish a recommendation from a decision, and can find the evidence they need. Other questions include how to show incomplete research, conflicting evidence, failures, and changes made to a draft. These are proposed validation questions, not completed research findings.
Project status
Design exploration shown in the supplied Figma file. The screens illustrate a proposed experience; they do not establish a working AI implementation, production release, or measured impact.