AI use cases in finance at German banks

In its report Risks in Focus, BaFin summarizes a survey by the European Banking Authority (EBA) among selected large EU banks. The most common areas of application were customer service, especially chatbots, then transaction monitoring, fraud detection, risk modeling and credit scoring. The same survey found generative AI at around 40 percent of the banks at the end of 2024 and at over 60 percent in the first quarter of 2025.

For the smaller German banks, the less significant institutions that BaFin and the Deutsche Bundesbank supervise directly, BaFin reports a narrower picture in the same text. AI there mainly writes texts, runs chatbots and helps risk management with fraud detection and the defense against cybercrime. It is rarely used to check creditworthiness or in trading. Savings banks and cooperative banks get much of their IT from their central providers, Finanz Informatik and Atruvia, so an AI tool chosen there reaches hundreds of banks.

The supervisors' view of bank AI is on AI in banking in Germany, insurers and asset managers are on AI in finance in Germany, and the high-risk rules are on the EU AI Act in financial services.

Customer service is the most common AI use in the EBA survey, and the chatbot is the part customers see. Commerzbank launched Ava, a banking avatar in its app built on Microsoft Azure AI. In the conversation, customers can order a new credit card, block or unblock a card or change limits. When a question gets too complex, Ava passes the customer to experts in the bank's customer center.

BaFin names customer chatbots as one of the AI uses it supervises under the German Act Implementing the European Artificial Intelligence Act. A banking chatbot is not a high-risk system under the EU AI Act, but customers must be told that they are talking to an AI. A chatbot that runs on a cloud model also brings a new ICT third-party provider into the bank, and DORA covers that contract.

Transaction monitoring and fraud detection come right after customer service in the EBA list, and BaFin names fraud detection as one of the main uses at smaller German banks. Section 25h of the German Banking Act (KWG) requires banks to run data processing systems that detect unusual transactions, so the duty exists whether the model inside works with fixed rules or with machine learning.

Atruvia, the IT provider of the Volksbanken and Raiffeisenbanken, checks transfers against a customer's usual pattern, such as the preferred channel and TAN method. Writing in Der Bank Blog, Andreas Hermann reported that the system detects and blocks criminal transactions correctly in four out of five cases (in German). Deutsche Bank is exploring a Google Cloud research agent for financial crime risk management.

The EU AI Act expressly keeps fraud detection out of its high-risk line on credit scoring. The rules that apply here are the KWG and the Money Laundering Act. Fraud prevention in Germany and regtech in Germany go deeper into both.

The first thing BaFin says about AI at smaller German banks is that it writes texts. The largest German example is S-KIPilot, the generative AI assistant that Finanz Informatik runs for the Sparkassen-Finanzgruppe in its own data centers, with no connection to a public cloud. According to the FI magazine (in German), about 147,000 people in the group use it, and in January 2026 they sent 1.95 million prompts. They summarized emails and long documents, prepared customer meetings and searched internal knowledge bases.

Commerzbank gave its staff cobaGPT (in German), a chatbot on Microsoft's Azure OpenAI Service. At Deutsche Bank, the research agent DB Lumina had around 5,000 users in Deutsche Bank Research, as Max Sommerfeld of Deutsche Bank and Crispin Velez of Google describe in their Google Cloud blog post on DB Lumina. It summarizes documents, extracts data, answers questions about uploaded files, drafts and translates.

Most of these assistants are minimal-risk under the EU AI Act, which asks mainly for AI literacy among staff. The heavier duty comes from DORA: BaFin published a guidance notice on ICT risks from AI that covers the whole life of a model, from data sourcing to shutdown, and the management of the cloud and model providers behind it.

Risk modeling is the next area in the EBA list. Banks have run statistical models for credit, market and liquidity risk for decades, and machine learning now sits next to them. The Bundesbank writes that banks use such methods to speed up processes, cut costs and make data available, and that the risks behind them fall under supervisory requirements. Together with BaFin it consulted the industry on machine learning in the risk models of Pillars 1 and 2, with explainability and data as the main questions.

A model that feeds a bank's capital requirements needs the supervisor's approval before it goes live, and a validation team that can explain what it does. The credit risk answer in the knowledge hub explains the basics.

Credit scoring is on the EBA list for large banks, but BaFin reports that smaller German banks rarely use AI to check creditworthiness. Regulation is one reason. Annex III of the EU AI Act lists AI that evaluates the creditworthiness of natural persons or sets their credit score as high-risk. A bank that deploys such a system must keep trained people in charge, monitor it and assess its effect on fundamental rights before first use. A bank that builds its own model also carries the provider duties, including documentation and bias testing.

The scoring of companies is outside that line, and corporate banking is where large German banks move first. Deutsche Bank is rolling out Google's Financial Research agent in its Corporate Bank, first with the teams serving German MidCorp clients, to prepare client briefings and financial analyses. BaFin's own principles for algorithms in decision-making ask for a named owner in the development phase and in the application phase of every such model.

Banks read and file large amounts of paper: identity documents for KYC, loan files, contracts and records of advisory talks. Commerzbank uses Google Gemini to produce MiFID-compliant documentation of advisory conversations with business clients (in German) and is widening its use of Google DocAI for document management. MiFID II requires a record of investment advice, so an AI that drafts the record saves advisors time, and the bank stays responsible for its content.

In KYC, the Money Laundering Act leaves the duty to identify a customer with the bank, even when a model reads the ID card or checks a company register extract. KYC in Germany covers the identification rules, and the knowledge hub explains KYC and AML.

Commerzbank uses Microsoft's AI programming assistant to modernize banking applications (in German). At fAInance, a conference by Sopra Steria and Fraunhofer IAIS for staff of financial institutions, AI-driven development and AI in software testing had their own stations. Code written with an assistant goes through the same change and test process as any other change to a bank's ICT systems, which DORA requires, and the assistant itself is an ICT service that belongs in the register of information.

Insurers use AI mostly to work faster and cut costs, BaFin reports, and rarely in pricing and risk assessment, the areas where the EU AI Act treats life and health insurance as high-risk. Allianz runs its Insurance Copilot for motor claims in Austria and has extended it to property claims. It gathers data from claims and contracts, reads documents and photos, flags discrepancies and drafts emails. The claims handler makes the final decision on cover and payout.

BaFin reports that smaller German banks rarely use AI in trading. Algorithmic trading falls under MiFID II whatever model drives it, and the firm must be able to stop an algorithm at any time.

The knowledge hub explains AI in trading and AI in wealth management. Automated portfolio management for private investors is on robo-advisors in Germany, and Finance Loop supported the AI & Market Data meetup in Paris on AI along the market data chain.

Finance Loop is the meeting place for people who build, buy and check AI at banks, insurers and asset managers. It connects the finance, IT and AI communities in Frankfurt and holds events in Munich, Berlin and Hamburg too.

Finance Loop presented fAInance, held in German and open only to staff of financial institutions, with stations on AI against financial crime, AI in governance, risk and compliance, an AI auditor and AI-driven development, and a closing talk on agentic AI in customer experience by Parloa. At the AI Week Frankfurt 2025 finance side event, run by Finance Loop with Frankfurt Main Finance and other partners, the talks covered deepfake awareness, AI as an investigator in anti-financial crime, and trusted AI and AI testing.

Finance Loop is a partner of AI in Finance Paris, organized by Finteda, with speakers from BNP Paribas, Deutsche Bank and Berenberg. It announced KI Exchange 2026 in Hamburg with AI governance and DORA compliance on the program, and it supports Claude Hacker House, a hands-on format on AI use cases in business and finance. With Frankfurt Data Science, Finance Loop has a cooperation on applied data science use cases and AI governance in financial institutions. Finance Loop member Fayssal El Mofatiche co-wrote Building AI Agents for Finance, which works through agents for fundamental analysis, deep research, trading, insurance and compliance.

In the EBA survey that BaFin summarizes, the most common AI uses at large EU banks were customer service with chatbots, transaction monitoring, fraud detection, risk modeling and credit scoring. At smaller German banks, AI mostly writes texts, runs chatbots and supports fraud detection.

In the EBA survey of selected large EU banks, around 40 percent used generative AI at the end of 2024 and over 60 percent in the first quarter of 2025. In Germany, about 147,000 people in the Sparkassen-Finanzgruppe use the S-KIPilot assistant, according to Finanz Informatik.

The model compares each payment with the customer's usual behavior, such as amount, channel and TAN method, and holds or blocks payments that do not fit. German banks must run such monitoring under section 25h KWG, and a person checks the flagged cases.

Large banks use AI to prepare credit analyses, mostly for corporate clients. For consumer loans, AI that scores a natural person is high-risk under the EU AI Act, and BaFin reports that smaller German banks rarely use AI for creditworthiness checks. The decision stays with trained staff.