Nitto Kogyo and IBM Japan Deploy Generative AI to Read Drawings for Distribution-Board Quotes
Electrical-equipment maker Nitto Kogyo has developed a multimodal generative-AI application that reads building and facility drawings and pulls out the circuit, equipment and special-note details needed to quote distribution boards, the company and IBM Japan said in a joint release on October 20. About 50 sales staff across 23 offices nationwide can now use it, with phased use in real quoting work starting this month.
Quoting a distribution board has long depended on experienced people. Every project arrives with drawings in a different format, and the estimator has to read regulations and product specifications off them before choosing devices and a circuit layout. The two companies spent about a year and a half organising that know-how, structuring the knowledge, deciding which steps people and AI each own, and setting up an operating model before the tool went live.
In the new workflow the AI reads the drawing and organises the data, while a salesperson checks and corrects the output. Both the AI extraction and the human corrections are stored as structured data, so quoting knowledge that used to sit with individual veterans can be shared and handed down. IBM Japan built the system on its integrated AI platform reference architecture and an agent framework drawn from its own Client Zero experience, and the companies say the design is not tied to any single AI model.
What it means in practice
The project sits alongside other Japanese efforts to point generative AI at dense technical paperwork, such as TEPCO’s test of AI checks on nuclear licensing documents and Ricoh’s multimodal model for Japanese document reasoning. The common thread is labour: construction and equipment firms face staff shortages and retiring specialists, so capturing judgement in data matters as much as speed.
Keeping a salesperson in the loop is also a practical safeguard. A misread breaker rating on a drawing becomes a wrong quote, and possibly a wrong panel on site, so the review step is where accuracy will be won or lost. The firms say they are also studying company-wide AI governance, security and architecture as use widens.
What to watch: how much reading accuracy improves as field feedback flows back into the tool, which additional drawing types it can handle, and whether Nitto Kogyo extends the approach into design, production and maintenance work as planned.