Customer Service Automation · Document Processing
Qonto wanted small-business customers to stop spending up to eight hours a month on repetitive financial admin such as transfers, invoicing, payroll, and cash-flow questions, but any agent touching bank accounts had to satisfy compliance, latency, and accuracy constraints.
Reported impact
Qonto made bank transfers 2x faster, cut invoice creation time 3x, sped payroll runs 5x, and saw customers trust agents with transfers as large as €100,000 while preserving final human approval for every action.
Qonto's AI Lab built in-app agents on the Claude Platform to act as both operators and analysts inside business banking workflows. Users upload invoices or payroll documents and chat with the agent instead of navigating the product manually. Complex steps run on Claude Opus 4.5 and Sonnet, while smaller steps go to Haiku, all through Amazon Bedrock so Qonto can use structured outputs, prompt caching, quota controls, and provider-level GDPR and EU AI Act compliance. The operator agent prepares transfers, drafts invoices, and converts payroll runs into bulk payments but always pauses for final customer approval before money moves. On the analysis side, agents answer questions over transaction data so customers can understand cash flow without exporting numbers into spreadsheets. Internally, Qonto's builders use Claude Code, Claude Cowork, and Claude Design to ship and refine the experience quickly.
Customers upload invoices, payslips, or ask a question in the Qonto app, the workflow routes each step to the right Claude model tier, the agent prepares transfers, invoice drafts, or payroll actions, and the customer gives the final approval before money or documents are sent.
Access runs through Amazon Bedrock, with GDPR and EU AI Act compliance handled at the provider level.
Cuts client invoice creation time 3X, with its AI agents drafting every invoice and the customer confirming before it's sent
The agents' complex missions run on Claude Opus 4.5 and Sonnet, while simpler steps go to smaller models, including Haiku.
We were quite surprised to see transfers of 100,000 euros with agents.
sends bank transfers 2X faster, with no manual beneficiary search, amount entry, or reference copying, then waits for the customer's final approval before the money moves
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
OpenAI Sample Oct 16, 2026 Itaú Itaú needed to make financial guidance simpler and more personalized for millions of customers while also scaling document-heavy internal workflows, all under the governance, observability, and economic controls… 100% of the customer base is the goal by the end of 2026 Customer DNA · IARA · OpenAI models 
Huawei Cloud Sample Archived · Apr 20, 2026 Postal Savings Bank of China Postal Savings Bank of China needed a scalable AI infrastructure and unified model platform so it could automate customer service, trading, document review, and risk workflows without creating disconnected AI pilots… RFQ volume exceeded 1.5 trillion yuan AI computing and data center · Huawei Cloud Stack · Youzhi large model 
Google Cloud Sample Archived · Jan 16, 2026 Hong Leong Bank Hong Leong Bank needed to replace slow manual banking operations and limited chatbot performance with a secure AI ecosystem that could support dynamic customer support, document-heavy trade operations, sanction… 30% efficiency gain Agent Development Kit (ADK) · Cloud SQL · Compute Engine 
