'Purchase Order AI-OCR (invox)' to Equip New Feature 'Reading AI Agent' from April 2026
Infomart and invox will add a LLM-powered 'Reading AI Agent' to their Purchase Order AI-OCR, enabling high-accuracy data conversion of complex forms through natural language instructions.
📋 Article Processing Timeline
- 📰 Published: April 8, 2026 at 20:10
- 🔍 Collected: April 8, 2026 at 11:30
- 🤖 AI Analyzed: April 20, 2026 at 20:05 (296h 35m after Collected)
Infomart Corporation (Headquarters: Minato-ku, Tokyo; President and CEO: Makoto Kimura, hereinafter "the Company"), which streamlines all business operations through the power of digital technology, will equip "Purchase Order AI-OCR (invox)"—provided in collaboration with invox Inc. (Headquarters: Shinjuku-ku, Tokyo; CEO: Akira Yokoi)—with a new feature, "Reading AI Agent", starting in April 2026.
With this feature, the AI learns user instructions to achieve "high-accuracy data conversion exactly as intended," which was previously difficult, thereby significantly reducing the man-hours and time required for order processing.
■ Background of the Feature Release
While digitalization is accelerating in corporate back-office operations, forms such as purchase orders received from business partners still have a mixture of proprietary layouts and formats for each company, and standardization has not progressed. Therefore, there have been cases where reading is difficult with conventional AI-OCR, creating a major barrier preventing the complete automation of order operations.
Especially in the school lunch industry, there are many purchase orders with extremely complex formats, such as specifying multiple delivery dates and locations for the same product. It is difficult for conventional AI-OCR to accurately recognize data from such purchase orders, and as a result, the time and effort required for the person in charge of receiving orders to visually check and manually enter/correct numerical values has become a heavy burden.
Therefore, to achieve further efficiency in order receiving operations, we will equip "Purchase Order AI-OCR (invox)" with the new feature "Reading AI Agent". The AI learns the user's instructions and realizes "high-accuracy data conversion as intended," significantly reducing the setup man-hours and time required for order processing and contributing to improved business productivity.
■ Feature Overview
[Capable of Reading Purchase Orders with Complex Formats, Such as Multiple Delivery Dates]
The "Reading AI Agent" allows users to directly instruct the AI on rules for reading purchase orders and data conversion methods. The AI understands the user's intent on "which items to extract and how" to convert it into data, making it possible to read purchase orders with complex formats such as the following:
- Cases where delivery dates and locations are split into multiple entries for the same product
- Cases where items are arranged irregularly vertically and horizontally
This eliminates the effort of "specifying the reading range (coordinate specification)" which was previously required for each form, significantly improving the accuracy and efficiency of data entry work.
Furthermore, this feature utilizes a Large Language Model (LLM), and a major characteristic is that detailed processing rules, such as "Delivery date column. If blank, make it the same value as the previous row," can be set using everyday language (natural language).
With this feature, the AI learns user instructions to achieve "high-accuracy data conversion exactly as intended," which was previously difficult, thereby significantly reducing the man-hours and time required for order processing.
■ Background of the Feature Release
While digitalization is accelerating in corporate back-office operations, forms such as purchase orders received from business partners still have a mixture of proprietary layouts and formats for each company, and standardization has not progressed. Therefore, there have been cases where reading is difficult with conventional AI-OCR, creating a major barrier preventing the complete automation of order operations.
Especially in the school lunch industry, there are many purchase orders with extremely complex formats, such as specifying multiple delivery dates and locations for the same product. It is difficult for conventional AI-OCR to accurately recognize data from such purchase orders, and as a result, the time and effort required for the person in charge of receiving orders to visually check and manually enter/correct numerical values has become a heavy burden.
Therefore, to achieve further efficiency in order receiving operations, we will equip "Purchase Order AI-OCR (invox)" with the new feature "Reading AI Agent". The AI learns the user's instructions and realizes "high-accuracy data conversion as intended," significantly reducing the setup man-hours and time required for order processing and contributing to improved business productivity.
■ Feature Overview
[Capable of Reading Purchase Orders with Complex Formats, Such as Multiple Delivery Dates]
The "Reading AI Agent" allows users to directly instruct the AI on rules for reading purchase orders and data conversion methods. The AI understands the user's intent on "which items to extract and how" to convert it into data, making it possible to read purchase orders with complex formats such as the following:
- Cases where delivery dates and locations are split into multiple entries for the same product
- Cases where items are arranged irregularly vertically and horizontally
This eliminates the effort of "specifying the reading range (coordinate specification)" which was previously required for each form, significantly improving the accuracy and efficiency of data entry work.
Furthermore, this feature utilizes a Large Language Model (LLM), and a major characteristic is that detailed processing rules, such as "Delivery date column. If blank, make it the same value as the previous row," can be set using everyday language (natural language).