KI in Banking-Chatbots und virtuellen Assistenten

Early banking bots were menu-driven or used intent classification. The system matched a message like 'what's my balance' to one of a few hundred predefined intents and ran a scripted flow. This design is predictable and easy to audit, which is why many large banks still rely on it for anything that touches accounts. Banks have adopted generative AI more cautiously. It often appears first in internal tools that help employees search procedures, or in bots that answer general questions from approved content. Well-known assistants include Bank of America's Erica and Capital One's Eno. Most of their work is routine: balances, transaction search, locking cards, payment reminders and routing people to the right team. Regulation shapes what these assistants may do. In 2023 the U.S. Consumer Financial Protection Bureau published an issue spotlight warning of three risks. Chatbots can give inaccurate information. They can trap customers in loops with no way to reach a person. And they can fail to recognize when a customer is asserting a legal right, such as disputing an electronic transfer error, which triggers obligations under laws like the Electronic Fund Transfer Act. Banks must also protect customer data under privacy rules, avoid unfair or deceptive practices, and keep records. Some jurisdictions require that people be told when they are talking to an AI system. The EU does so under its AI Act. Common failures include misunderstanding unusual requests, giving confident but wrong answers about fees or policies, and refusing to hand off to a person. Chatbots can also be attacked. Someone may try to trick a generative assistant into revealing information or taking an action. A common misconception is that a chatbot can do anything a teller can. In practice, sensitive actions usually require strong authentication and fixed, rule-based back-end processes. The language model acts only as the front end.