AI in KYC: How Agents Help DZ BANK at the Customer Interface
DZ BANK is using artificial intelligence to reduce routine work. This frees up time for employees to focus on more demanding tasks. The bank sees the greatest benefit in embedding the technology into its core processes. One example is the KYC ("Know Your Customer") process, in which banks verify who their customers are and who ultimately stands behind them. DZ BANK is currently testing an AI agent that takes over the preparatory work and is scheduled to go live in the coming months. Florian Mielke, Head of AI Office, explains how it works and why a human still makes the final decision.
Mr. Mielke, why is DZ BANK using AI in the KYC process?
The process is a good fit because it is document-intensive, recurring and data-driven. At the same time, it is demanding both in terms of subject matter and regulation. The necessary information is scattered across commercial register extracts, shareholder lists and organization charts. With multi-tier shareholdings or foreign companies, consolidating it is especially time-consuming. This step also has a clear beginning and a clear end: Documents go in, an ownership structure comes out. That makes it easy to measure how reliably the agent works.
How does the AI agent work, and does the technology make the final decision?
No, the decision stays with humans. The agent reads the documents, organizes the information and checks it for plausibility. The result is an overview of who holds a stake in whom, and every statement in it points to its source. Subject-matter experts review the result, add missing information and make the decision. The more complex a case, the more human involvement it requires.
What is DZ BANK learning from the project?
Demos are not enough to scale AI. Building a first agent is quick, but the real work begins afterward: Accuracy has to reach a reliable level and stay there. That is an ongoing task. Nor is it enough to simply add agents to existing processes, because you may end up automating their weaknesses. Clear processes, control points and roles are needed to make the technology deliver lasting, dependable value.
A detailed guest article by Florian Mielke on this topic is available here.
Author: Robert Leonhardt, Press Spokesperson at DZ BANK
