Lumıary
Company reportedLive since Apr 2026Added Oct 2026

Qonto

Qonto builds Claude-powered agents to handle small-business payments and invoicing

Qonto, a European business bank, built two AI agents into its app to prepare payroll transfers, draft invoices and answer questions about customers' money. Qonto reports transfers 2X faster, client invoices 3X less time and payroll 5X quicker, with the customer confirming every critical action.

Raise productivityImprove customer experience

Key figures

Copied exactly from the sources. Copy any figure with its citation.

2

X faster

Speed of sending bank transfers with the agents, as reported by Qonto1

3

X less time

Time to create client invoices with the agents drafting them1

5

X quicker

Speed of running payroll with the agents1

as large as €100,000

Largest transfers customers have delegated to the agents1

What it took

CostNot disclosed
TeamA 15-person AI Lab1
TimelineProject started November 2025; first agent released six weeks later to several thousand beta customers1

The problem

Qonto's customers are mostly solo entrepreneurs, freelancers, new companies and small teams. Few have a finance team, so they do repetitive money admin by hand. An owner running payroll typically made one manual transfer per pay slip, usually 15 to 20 a month, and typed client invoices into the platform one at a time.

What they built

Two AI agents inside the Qonto app. An operator agent prepares the repetitive work: payroll and supplier transfers, and client invoices drafted from a list of clients. An analyst agent answers questions about past and present transactions. Qonto also offers a connector that lets customers use their own Qonto data from Claude directly.

How they did it

Qonto's 15-person AI Lab started in November 2025 and released a first agent six weeks later to several thousand beta customers. Complex agent tasks run on Claude Opus 4.5 and Sonnet, simpler steps go to smaller models including Haiku, and each step is routed by speed, quality and cost. Access runs through Amazon Bedrock. Qonto re-runs evaluations on each new model release. Every critical action needs the customer's confirmation before it executes.

Results

Qonto reports that bank transfers are sent 2X faster, client invoices take 3X less time to create and payroll runs 5X quicker, with a 15-to-20-slip run turned into one bulk transfer. Customers have delegated transfers as large as €100,000 to the agents, and one generated 500-plus monthly invoices in a single batch. All figures are Qonto's own, shared through the vendor's story.

All reported figures

MeasureFigureCitation
Speed of sending bank transfers with the agents, as reported by Qonto2X faster1
Time to create client invoices with the agents drafting them3X less time1
Speed of running payroll with the agents5X quicker1
Largest transfers customers have delegated to the agentsas large as €100,0001
Invoices one customer generated in a single batch each month500-plus1

Lessons for your project

  1. Keep the human approval step. Qonto's agents prepare transfers and invoices, and the customer confirms before money moves or an invoice is sent.

  2. Pick tasks where volume, not difficulty, is the pain. Payroll, with 15 to 20 transfers a month, became one bulk transfer.

  3. Involve legal, risk and security from day one. Qonto credits its co-building of the solution with these teams for the pace of the first release.

  4. Route work to different model sizes by latency, quality and cost, and re-run evaluations on every new model release, as Qonto does.

Caveats

  • The headline speed figures come from Qonto and from Anthropic's customer story. Neither gives a measurement method, and no independent source confirms them.
  • The Anthropic story is undated. It places the project start in November 2025 and the first agent six weeks later. Qonto's public launch of two agents for all customers is dated 26 April 2026 by FF News, so the deployment date shown is the public launch and the beta started earlier.
  • The customer story says customers spend up to 8 hours a month on financial admin, citing Forrester analysis for Qonto's customers. The FF News article says up to 8 hours a week. That figure is not used here.
  • Only the Anthropic story names the Claude models. Qonto's own page and the FF News article do not name any model.
  • These agents serve Qonto's customers inside its product. The sources do not report back-office savings at Qonto itself.

Sources

  3. Qonto IA : vos agents IA pour la finance d'entreprise

    Qonto · Undated · Company blog · qonto.com

Summarized in our own words from the sources above. Spotted an error? See corrections.

Related cases