Intuit Supports Accounting and Tax with AI. What is its proprietary OS "GenOS"?

Which Company's Case Study?

This is a case study of Intuit, a US financial technology company. The company provides accounting software "QuickBooks" for small and medium-sized businesses and sole proprietors, tax filing software "TurboTax", and email marketing service "Mailchimp", among others. Leveraging vast financial data and expertise accumulated over many years, Intuit is company-wide promoting the use of generative AI to support users' financial success.

Problems Intuit Wanted to Solve

Small and medium-sized businesses and sole proprietors spend a lot of time on a wide range of complex tasks, including daily transaction entry, invoice creation, tax filing, and marketing activities. Many of these tasks require specialized knowledge, making them prone to errors or difficult to obtain timely insights necessary for business decisions. Intuit recognized the need for AI that could provide expert-like assistance, enabling these users to manage their finances more easily and confidently.

How AI Was Used

Intuit developed its proprietary generative AI operating system, "GenOS" (Generative AI Operating System). This platform serves as the brain for "Intuit Assist", the AI assistant integrated into the company's products (such as QuickBooks and TurboTax). GenOS is centered around Financial Large Language Models (Financial LLMs) custom-trained on vast financial data accumulated over decades. This enables more accurate and contextually relevant answers to specialized questions in areas like accounting and tax. It also leverages Retrieval-Augmented Generation (RAG) technology to enhance answer accuracy by referencing the latest tax laws and help articles. Users can simply speak to Intuit Assist in natural language to get answers to questions about their financial situation, obtain personalized insights, and automate tasks such as invoice creation.

Anticipated Architecture

The following is an anticipated architecture estimated from publicly available information and common configurations. It does not definitively state the actual internal structure.

• ▸Input Data: Natural language questions or commands entered by users into the "Intuit Assist" UI. Contextual data within each product, such as accounting data on QuickBooks, tax information from TurboTax, and campaign data from Mailchimp.
• ▸AI Processing / Model or Search Layer: The proprietary generative AI platform "GenOS" plays a central role. This platform is built on AWS. GenOS consists of components such as "GenRuntime" for real-time selection of the optimal LLM, and "GenStudio" for developers to quickly build and experiment with AI functions. Internally, it is believed to utilize not only custom-trained financial LLMs but also multiple external LLMs like Anthropic's Claude and Google's Gemini.
• ▸Output to Business Systems: AI-generated answers and insights are displayed on the "Intuit Assist" chat screen. It also directly executes functions of each application, such as automatic report generation, invoice creation, and drafting marketing emails.
• ▸Monitoring and Governance: Given the highly sensitive nature of financial data, a strict governance system is in place to ensure data privacy, security, and the accuracy of AI responses. Alongside AI automation, review and support by human experts are also provided.

Implementation Effects and Key Points to Note

• ▸Observable Effects and Changes: Through interaction with the AI assistant, users can complete complex accounting processes and tax filings with less effort and more confidence. By leveraging AI, Intuit performs over 65 billion machine learning predictions per day and generates 810 million AI-driven customer interactions annually.
• ▸Key Points for Readers to Note: Instead of simply using a general-purpose LLM, Intuit has built custom LLMs with vast amounts of its own domain data and developed its unique "GenOS" platform. This allows for high accuracy and reliability even for industry-specific complex tasks, establishing a strong competitive advantage.
• ▸Caution: In fields where errors are unacceptable, such as finance and tax, mechanisms to ensure the accuracy, up-to-dateness, and safety of AI responses are essential. Intuit has established guardrails for responsible AI development at the platform level, and such initiatives are key to successful implementation.

What Japanese Companies Can Learn

Intuit's case demonstrates that proprietary data and domain knowledge accumulated by a company over many years become its most valuable assets in the era of generative AI. While not all companies can develop their own LLMs, approaches like RAG, which integrate a company's manuals, past inquiry histories, and operational data as a knowledge base for AI, are applicable to many businesses. Starting with the creation of a small-scale AI assistant specialized in specific tasks, aiming for operational efficiency and improved customer experience, would be a realistic first step.

Today's AI News

Reference Links

Find Related AI Adoption Case Studies