Twin AI Agent for Qonto Invoice Retrieval

Twin Launches AI Automation Agent with Qonto

Emerging from stealth in January 2024, Twin entered a landscape where AI agents were largely conceptual. Currently, the Paris-based firm is unveiling an automation agent developed in collaboration with Qonto, a fintech company providing business banking solutions to over 500,000 clients throughout Europe.

Existing Automation Solutions

Several methods already exist for automating routine processes. Some organizations utilize API-driven, no-code, or low-code automation tools, such as Zapier. Others depend on Robotic Process Automation (RPA) software, like UiPath.

Twin's Approach to Automation

Twin, with its team of nine, proposes a more streamlined approach to automation. This involves leveraging artificial intelligence and computer usage models, as the company believes this offers a significant advantage.

Invoice Operator: A Practical Application

Twin’s initial product, Invoice Operator, created for Qonto, exemplifies the benefits of employing artificial intelligence. Qonto processes millions of invoices monthly, and its customers dedicate considerable time to collecting and uploading these invoices.

How Invoice Operator Works

Over the last three months, Twin developed an automated invoice retrieval system to accelerate this process. Upon initiating Invoice Operator, the system first identifies transactions lacking invoices.

It then displays a list of services requiring access for invoice downloads alongside a browser window illustrating the agent’s activities.

Secure Credential Input

If service login is necessary for invoice download, the browser will pause, prompting manual credential entry. Once provided, a button allows the agent to resume its operation.

Automated Invoice Processing

Subsequently, Twin’s Invoice Operator automatically locates past transactions, downloads the corresponding invoices, and attaches the PDF files to the transactions within the user’s Qonto account.

Scaling Automation with AI

“Scaling to the level of Qonto necessitates accommodating a vast and diverse range of services – thousands, and soon hundreds of thousands, utilized by everyone,” explained Twin co-founder and CEO Hugo Mercier during a product demonstration.

“Achieving this with RPA would be impractical, requiring a unique script for each website and constant script modification with every website update,” he further stated.

Twin vs. API-Based Automation

Mercier noted that it took Zapier a decade to integrate support for 8,000 applications on its platform. In contrast, Twin’s Invoice Operator already supports thousands of applications within just a few months of development.

Underlying Technology

Twin operates a server-based Chromium web browser. The startup utilizes OpenAI’s CUA (computer-using agent) model, and was among the initial 15 companies to beta test CUA.

CUA and OpenAI’s Operator

CUA also powers OpenAI’s Operator, a prosumer product enabling users to initiate actions through prompts. Twin believes that agents performing web browsing should be more user-friendly and offer improved performance.

Simplified User Experience

“We prioritized a remarkably simple user experience, targeting even those less familiar with technology. Users need not prompt or configure anything; they simply log in, launch the agent, and it autonomously locates the invoices,” Mercier clarified.

Future Applications for B2B Agents

Following invoice retrieval, Twin envisions numerous industries benefiting from agentic applications in the B2B sector. These could include automated order management for e-commerce, catalog classification for marketplaces, or information retrieval for call center personnel.

The Future of AI Agents

Twin anticipates a future where AI agents are more affordable, efficient, and precise across a broad spectrum of tasks. The company’s next step is to transform the core agent platform powering Invoice Operator into a product accessible to developers for use in their own applications.