Case Study: Agentic KYC & AML Onboarding for Commercial Banking

Banking

Agentic KYC & AML Onboarding for Commercial Banking

The Challenge: Twenty-Three Days to Open a Business Account

A regional commercial bank took an average of 23 days to open an account for a mid-sized business, and some prospective clients went elsewhere before the process finished. KYC analysts spent that time collecting registry filings and tracing beneficial ownership through layered holding structures, then ran sanctions and adverse media checks and wrote the risk assessment by hand. Files differed from one analyst to the next, and a regulatory review had found gaps in how ownership conclusions were evidenced. The bank had tried rules-based automation, which could not handle the variety of corporate documents and ownership structures its analysts saw.

Our Solution: A Governed Multi-Agent Onboarding Workflow

Lydatum rebuilt the onboarding process around four AI agents, each responsible for one part of the KYC file, with an analyst reviewing and approving every file before a decision is made:

• Document & Entity Agent: Collected registry filings and client-supplied records, then matched each legal entity against internal and third-party reference data.
• Beneficial Ownership Agent: Built the ownership graph for each client and traced control through intermediate entities to the individuals behind them. Where the evidence was incomplete or contradictory, it flagged the structure for the analyst.
• Screening Agent: Ran sanctions and adverse media checks on every entity and individual, including politically exposed persons, and passed on only the hits that needed a judgment call, each with its reasoning.
• Risk Narrative Agent: Drafted the KYC risk assessment in the bank's own template, with every statement linked to the document that supported it.

Each agent worked through a controlled tool layer with its own permissions, and the system logged every agent action and analyst decision. No file could be approved without analyst sign-off, and higher-risk clients went to compliance officers automatically. The system ran in shadow mode beside the existing team for six weeks before go-live, which gave the bank's compliance and model risk teams their own evidence of how it performed.

The Impact: Faster Onboarding With a Stronger Evidence Trail

Four months after go-live, the agents were preparing the full KYC file for 85% of new commercial clients, and analysts spent their time reviewing conclusions.

Average onboarding time fell from 23 days to just over 7, and each analyst handled three times as many cases, which cleared the backlog within a quarter. The bank's next internal audit sampled files across the team and found ownership conclusions evidenced in every one.

Services Delivered: Agentic Workflow Automation, Multi-Agent Systems & Orchestration, Agent Governance & Controls, Context Engineering & Enterprise Knowledge

Technologies Used: LangGraph, Anthropic Claude on Amazon Bedrock, Model Context Protocol (MCP), Amazon Textract, Neo4j, Snowflake, Python