Generative AI in finance

Generative AI is AI that produces new text, code or images from a prompt, built on large language models (LLMs). Older AI in finance classifies and predicts: it scores a loan or flags a card payment. Generative AI drafts the email to the customer, summarizes a long credit file and answers an employee's question about an internal policy. In its report Risks in Focus, BaFin summarizes a European Banking Authority survey of large EU banks in which the share that used generative AI rose from around 40 percent to over 60 percent within a few months. In ESMA's survey of EU securities markets, generative AI was behind most reported AI use cases.

The trade body UK Finance mapped where firms get the most value in its report Generative AI in Action: customer engagement and marketing, knowledge management and information retrieval, software development and data management, email and workflow processing, fraud and financial crime, analysis of legal, contract and compliance texts, and meeting productivity. The same study found about 70 percent of financial firms still piloting generative AI, while 91 percent already used predictive AI for tasks such as fraud detection.

In Germany the biggest deployment runs at the savings banks. Finanz Informatik operates S-KIPilot for the Sparkassen-Finanzgruppe in its own data centers, with no connection to a public cloud, and about 147,000 people in the group use it, according to the FI magazine (in German). DekaBank gave its staff DekaGPT 2.0, which searches several long documents at once, and Handelszeitung (in German) reports almost 1,800 users of its internal chatbots. ING Germany uses generative AI in customer service and grounds the answers in its own documents through retrieval-augmented generation, as silicon.de (in German) describes.

Retrieval-augmented generation, or RAG, is the pattern behind most of these tools. The model does not answer from memory; it first retrieves passages from approved internal sources, such as product sheets or policy manuals, and writes its answer from them. That makes answers traceable and cuts down on invented facts, which matters when the text goes to a customer or into a credit file. The AI use cases in finance page covers the wider list of AI tools at German banks.

Language models can "hallucinate": they write fluent text that is wrong. ESMA warns that in advice or portfolio reports this can mislead clients, and every bank tool above keeps a person in the loop for that reason. The Financial Stability Board looks at the system as a whole. Its report on the financial stability implications of AI names third-party dependencies and the concentration of a few model and cloud providers, correlated behavior across markets, cyber risk, and model risk with data quality and governance as the vulnerabilities that stand out.

For a German bank, the concentration point is a legal duty under DORA. BaFin's guidance on ICT risks from AI covers the whole life of a model, from data sourcing to shutdown, and the management of the cloud and model providers behind it. A staff chatbot is minimal-risk under the EU AI Act, which mainly asks for AI literacy among the staff who use it; the companies that build the underlying general-purpose models carry their own documentation and transparency duties under the act.

Finance Loop is the meeting place for the people who roll out generative AI in finance: IT architects, compliance officers, product owners and the business teams that use the tools. It connects the finance, IT and AI communities in Germany, Austria and Switzerland, with events in Frankfurt, Munich, Berlin and Hamburg.

With Finteda and Frankfurt Data Science, Finance Loop supports Claude Hacker House, where developers and finance people build working prototypes with Claude, and it offered selected members a place in Anthropic's Claude Certified Architect Foundations certification. Finance Loop presented fAInance, where the keynote by Fraunhofer IAIS went from the LLM copilot to autonomous systems, and announced KI Exchange in Hamburg, with AI governance and DORA compliance on the program. At the AI Week Frankfurt 2025 finance side event, run by Finance Loop with Frankfurt Main Finance and other partners, Jasper Albrecht of revel8 spoke on deepfake awareness.

Drafting and summarizing text, internal search and assistants for staff, customer service chat, code generation, and the analysis of contracts and compliance documents. In fraud and anti-money laundering work, it helps investigators write up cases, while predictive models still do the detection.

Traditional AI in finance predicts or classifies, such as the probability that a borrower defaults. Generative AI produces new content, such as a draft reply or a summary of a credit file. Many banks combine both: a model flags a case, and a language model writes the first draft of the report.

Claude Hacker House is a hands-on evening format in Frankfurt for people from finance and tech. For a degree, Frankfurt School of Finance & Management offers a part-time Master of Artificial Intelligence & Data Science, and the hub answer on AI in finance explains the basics.