Banking: documented AI cases, with their source
This page includes publicly described uses of AI in a bank, neobank or credit institution. A case enters if the organization, title or summary names this profession — not if the word "bank" appears only in a personal approach.
The figures quoted on the sheets are those of the source. A declaration of publisher does not have the same value as a statement of establishment; the level of proof on each card reminds him.
This selection has 23 spotted cases. 21 concern an organisation in France or the French-speaking world. Proof levels: 16 « Testimony published by an identified source » and 7 « Statement by the tool’s publisher ». Most cited tools: Claude (3), Dust (2) and Mistral (2).
Recurring caveats
- 10 cards out of 23 mention verification of figures, sources or extracts.
- 8 cards out of 23 mention human rereading or validation.
- 5 cards out of 23 mention a legal, tax or compliance framework.
- 4 cards out of 23 mention personal data, a trade secret or the.
- 2 cards out of 23 mention what the tool can send or create on its own.
Count taken from the “vigilance” field of the cards on this page, without rephrasing the caveats.
23 spotted cases
- Control the real estate credit files and shorten the grant deadline
Since November 2024, BNP Paribas has been deploying in France an intelligent processing of documents that reads the pieces of a real estate credit file, checks its consistency and saves about five days on the award journey.
- Assist advisors in interviewing the bank's knowledge base
On the SG network in France, a documentary search tool enriched by the general AI answers the advisors' questions from the internal knowledge base, referring to the source documents.
- Produce customized market research reports from hundreds of sources
Crédit Agricole CIB has integrated an AI platform that links billions of internal and external data to help its market research teams write thematic reports tailored to each client's profile.
- Analyze e-mailing campaigns of a bank with a copilote that checks its own numbers
For La Banque Postale, Numberly has deployed Smart Insights, an AI architecture that extracts indicators from an e-mail campaign, writes a first analysis and has the numbers audited by a second agent before human editing.
- Help anti-fraud analysts and a neobank's customer service
At Nickel, a subsidiary of BNP Paribas, the general AI is already in production to prioritise fraud alerts and assist customer service: the neobank estimates the savings already made on this side at EUR 1 million.
- Piloting a tertiary building by communicating with its data on a pilot site
BNP Paribas Real Estate is experimenting with Willow Copilot, an assistant who questions real-time building data; the first deployment, presented as a pilot, takes place at METAL 57, its seat in Boulogne-Billancourt.
- Online on Concept, Dust produces the first marketing content bases and adapts them from one market to another
Marion de Robillard, world marketing and communication director of Qonto, connected Dust (RAG) to the Internal Concept, then marketing bots to produce first content bases and adapt them from one country to another — from 4 to 8 markets.
- Respond to bank customers with open source RAG chatbot supervised by experts
Crédit Mutuel Arkéa built its "Tybot" client chatbot on Tock, an open source conversational toolbox launched by SNCF, looking for answers from the customer knowledge base and product sheets.
- Framework the deployment of hundreds of trades-built agents
Bpifrance opened a codeless agent platform to 500 pioneer collaborators, after a four-month experiment and the establishment of a governance framework with Artefact.
- Transfer a frozen ios script banking chatbot to generated answers, in three languages
The Belfius bank replaced a conversational robot in scriptios with an assistant based on a large language model, available in Dutch, French and English.
- Prepare transfers, invoices and pay from the bank application, final validation by the customer
Qonto, a neobank of small and independent companies active in eight European markets, has integrated agents who prepare repetitive operations into its application; the customer always keeps the final validation.
- Compare dozens of internal directives to regulations in a matter of hours
The Swiss private bank Pictet has opened a code agent and an office automation agent to approximately 700 employees; The compliance team is used for an analysis of regulatory deviations previously considered out of reach.
- Instruct card claims and sort claim letters
The online bank N26, which is active in 24 European markets, has integrated an AI model into more than 15 internal applications, from card payment disputes to financial crime analysis.
- Preparing calls from remote banking advisors with an AI assistant
CaixaBank, Spain's first bank, offers more than 300 remote support advisors an assistant who prepares calls and answers questions about products.
- Transforming product marketing document templates into skills for AI
A Qonto product marketing manager converted the templates of his trade (communication notes, change logs, launch kits) into thirteen AI skills that draw from Notion and Slack.
- Accompanying employees in a transition to AI rather than replacing them
Steve Anavi, one of Qonto's leaders, describes how the neobank, 1,600 employees, organizes the requalification of its teams: management and budget from above, experiments proposed by the teams themselves.
- Answer an analytical agent from validated indicator definitions
Qonto opened access to its data warehouse via Claude agents, found false but plausible figures, and then rebuilt a layer of validated definitions on which the agents rely.
- Answer practical questions from clients in the Apps Bank populaire and Caisse d'Epargne
Yves Tyrode, Digital DG and BPCE payments, said that the AI assistant for Banque populaire and Caisse d'Epargne applications generates more than 5,000 conversations a day, and that more than one in two leads to an action proposed by AI, often for a practical question.
- Reread its code, generate tests and documentation at a BNP Paribas CIB developer
Chloé Stephan, a full-stack developer at BNP Paribas CIB, sends his code to the Developer Integrated Assistant at IDE: "You are a senior developer. Critique my code. » The tool also automates unit tests, documentation and tickets Jira; the advertised gain is 8 to 15%.
- Dealing with simple 24 hour requests on Centric and Neolink with Noa, at BNP Paribas CIB
Jean-Marc Friess, Chief Digital Services Officer of BNP Paribas CIB, describes Noa, virtual agent of the Centric and Neolink portals (100 000 users): intention by a model, deterministic answer via internal API, otherwise ticket synthesized to the right team.
- Gathering changes, past incidents and resolution paths for P1 incidents at BNP Paribas' DSI
Marc Camus, DSI Group of BNP Paribas, has deployed an internal incident management assistant: in a significant incident, he aggregates changes in resources, precedents in the internal tool and already identified resolution paths. About 80% of criticality 1 events are already managed with this assistant.
- Support outreach via a chatbot Teams plugged into the internal database, at BNP Paribas' DSI
Marc Camus, DSI group of BNP Paribas, is testing in the only DSI group a chatbot in Teams who questions the internal knowledge base for outreach support. The next step is to connect it to the contact centre if the user does not solve alone.
- Ask the whole body of research, summarize and score the feeling, with Alix at BNP Paribas CIB
Goulven Charles, Chief Data Officer of BNP Paribas CIB, describes Alix, conversational assistant of Data Lab Global Markets: natural language on the entire research corpus, abstracts, extraction, translation, and sentiment scores (growth, central banks, inflation) on about 100 publications per day.