Accounting AI Agent Co-Developed by Ajinomoto Financial Solutions and Fast Accounting Enters Production Operation
Ajinomoto Financial Solutions and Fast Accounting have co-developed an 'Accounting AI Agent' that autonomously handles expense approvals. Deployed in production, it achieves a 93.3% accuracy rate and is expected to save 10,000 hours annually.
📋 Article Processing Timeline
- 📰 Published: April 24, 2026 at 19:00
- 🔍 Collected: April 24, 2026 at 10:31
- 🤖 AI Analyzed: April 25, 2026 at 02:57 (16h 25m after Collected)
Ajinomoto Financial Solutions Inc. (Headquarters: Chuo-ku, Tokyo; President: Yasushi Takagi; hereinafter AFS), a wholly-owned subsidiary of Ajinomoto Co., Inc. handling financial and accounting operations for domestic group companies, and Fast Accounting Co., Ltd. (Headquarters: Minato-ku, Tokyo; President: Keitaro Mori; hereinafter Fast Accounting), which aims to achieve accounting singularity, have been jointly developing an accounting AI agent equipped with specialized accounting and tax knowledge.
We are pleased to announce that this AI agent has begun production operation ahead of schedule at AFS.
In this project, we built a system where AI autonomously handles the entire workflow—from logging in to confirming, approving, or judging whether to return applications—for expense settlement approval tasks that were conventionally done manually. As a result, we confirmed that operational efficiency can be vastly improved while maintaining quality.
We position this initiative as a critical step toward transforming accounting operations through AI.
■ Background of the Project
While centralizing domestic financial accounting operations for the Ajinomoto Group, AFS faced the following challenges:
- Tens of thousands of hours spent annually checking expense settlements and invoice processing
- Personalization of judgment tasks requiring accounting and tax knowledge
- Difficulty in passing down know-how
- Limits to quality maintenance and sustainability even when utilizing BPO (Business Process Outsourcing)*.
*BPO: Outsourcing part of business processes in corporate activities to specialized vendors.
To address these issues, Fast Accounting has long been researching accounting-specialized AI language models designed and developed in-house.
The two companies collaborated to launch this project, aiming to build a system utilizing an accounting AI agent incorporating AI technology to perform speedy and sustainable accounting judgments while maintaining work quality. The goal was an 'AI that takes on accounting judgment itself,' going beyond mere RPA or existing generative AI usage.
■ Construction of the Accounting AI Agent
Specifically, the accounting approval tasks for multiple companies entrusted to AFS were reconstructed on the premise of AI utilization. The accounting-specialized AI technology provided by Fast Accounting was combined with their long-cultivated AI OCR, voucher processing, and business automation technologies (such as Remota/Robota), embedding the Ajinomoto Group's business rules and knowledge. As a result, the series of steps required for accounting approval (from login to judgment/approval) was automated, enabling flexible judgments based on accounting knowledge rather than simple rules.
Moreover, unlike RPA, it does not rely on screen items or conditions, offering versatility and the potential for horizontal deployment to other companies and tasks. This achieves automation with reproducibility and effectiveness while keeping verification costs low.
Through this, accounting judgments for expense settlements, which previously took 4 to 5 minutes per item, can be automated by matching application data with images. In accounting sites processing around 10,000 vouchers monthly, an annual reduction effect of approximately 10,000 hours is expected. After going live for internal expense settlement operations at AFS, it will be expanded to the approval operations of other entrusted companies.
Furthermore, in the future, we aim to automate accounting judgments in invoice processing operations as well, such as matching application data with images, judging prepaid expenses, judging capital expenditures, handling withholding taxes, and judging lease transactions.
Flow Image of the AI Agent in Early Operation at AFS:
AI makes judgments, while humans focus on internal control assurances like ex-post sample checks.
■ Proprietary Technology Complementing the Limits of Generative AI
While Large Language Models (LLMs) have high generative capabilities, they face the challenge of 'hallucination'—plausibly outputting factually incorrect content. In accounting operations, this can be a barrier to establishing AI utilization. It is said that even as the accuracy of LLMs themselves improves, completely eliminating this problem is difficult.
Because Fast Accounting has internally developed AI specifically for accounting, we tackled this issue head-on. Consequently, rather than just increasing the output of the LLM, we prioritized completing tasks in a trustworthy manner. We built the AI agent by designing business prerequisites, procedures, and confirmation points (acceptance criteria) before applying the AI. By incorporating various accumulated technologies, an accounting AI agent capable of fully automating the series of accounting approval steps was realized.
■ 93.3% Accuracy Rate
To verify the effectiveness of the 'Accounting AI Agent,' which combines the business logic in the accounting/finance domain cultivated by Fast Accounting with an LLM, we tested it for the three items of
We are pleased to announce that this AI agent has begun production operation ahead of schedule at AFS.
In this project, we built a system where AI autonomously handles the entire workflow—from logging in to confirming, approving, or judging whether to return applications—for expense settlement approval tasks that were conventionally done manually. As a result, we confirmed that operational efficiency can be vastly improved while maintaining quality.
We position this initiative as a critical step toward transforming accounting operations through AI.
■ Background of the Project
While centralizing domestic financial accounting operations for the Ajinomoto Group, AFS faced the following challenges:
- Tens of thousands of hours spent annually checking expense settlements and invoice processing
- Personalization of judgment tasks requiring accounting and tax knowledge
- Difficulty in passing down know-how
- Limits to quality maintenance and sustainability even when utilizing BPO (Business Process Outsourcing)*.
*BPO: Outsourcing part of business processes in corporate activities to specialized vendors.
To address these issues, Fast Accounting has long been researching accounting-specialized AI language models designed and developed in-house.
The two companies collaborated to launch this project, aiming to build a system utilizing an accounting AI agent incorporating AI technology to perform speedy and sustainable accounting judgments while maintaining work quality. The goal was an 'AI that takes on accounting judgment itself,' going beyond mere RPA or existing generative AI usage.
■ Construction of the Accounting AI Agent
Specifically, the accounting approval tasks for multiple companies entrusted to AFS were reconstructed on the premise of AI utilization. The accounting-specialized AI technology provided by Fast Accounting was combined with their long-cultivated AI OCR, voucher processing, and business automation technologies (such as Remota/Robota), embedding the Ajinomoto Group's business rules and knowledge. As a result, the series of steps required for accounting approval (from login to judgment/approval) was automated, enabling flexible judgments based on accounting knowledge rather than simple rules.
Moreover, unlike RPA, it does not rely on screen items or conditions, offering versatility and the potential for horizontal deployment to other companies and tasks. This achieves automation with reproducibility and effectiveness while keeping verification costs low.
Through this, accounting judgments for expense settlements, which previously took 4 to 5 minutes per item, can be automated by matching application data with images. In accounting sites processing around 10,000 vouchers monthly, an annual reduction effect of approximately 10,000 hours is expected. After going live for internal expense settlement operations at AFS, it will be expanded to the approval operations of other entrusted companies.
Furthermore, in the future, we aim to automate accounting judgments in invoice processing operations as well, such as matching application data with images, judging prepaid expenses, judging capital expenditures, handling withholding taxes, and judging lease transactions.
Flow Image of the AI Agent in Early Operation at AFS:
AI makes judgments, while humans focus on internal control assurances like ex-post sample checks.
■ Proprietary Technology Complementing the Limits of Generative AI
While Large Language Models (LLMs) have high generative capabilities, they face the challenge of 'hallucination'—plausibly outputting factually incorrect content. In accounting operations, this can be a barrier to establishing AI utilization. It is said that even as the accuracy of LLMs themselves improves, completely eliminating this problem is difficult.
Because Fast Accounting has internally developed AI specifically for accounting, we tackled this issue head-on. Consequently, rather than just increasing the output of the LLM, we prioritized completing tasks in a trustworthy manner. We built the AI agent by designing business prerequisites, procedures, and confirmation points (acceptance criteria) before applying the AI. By incorporating various accumulated technologies, an accounting AI agent capable of fully automating the series of accounting approval steps was realized.
■ 93.3% Accuracy Rate
To verify the effectiveness of the 'Accounting AI Agent,' which combines the business logic in the accounting/finance domain cultivated by Fast Accounting with an LLM, we tested it for the three items of