What Changed: 22 Hours to 3 Hours per Patent Search
In April 2026, NEC Corporation launched the commercial rollout of its AI-powered Intellectual Property DX Service, combining generative AI with RAG (Retrieval-Augmented Generation) technology. Internal validation demonstrated a reduction in per-patent search time from approximately 22 hours to approximately 3 hours — a maximum efficiency gain of 94%. NEC has set a target of ¥300 billion in IP DX business revenue by the end of fiscal year 2030.
Traditional patent search required engineers and IP professionals to manually enter keywords into patent databases (J-PlatPat, Espacenet, Derwent, etc.), read dozens to hundreds of retrieved documents, and assess their technical relevance. Prior-art searches conducted before filing a patent application, and freedom-to-operate analyses to verify that a competitor’s product does not infringe one’s own patents, are particularly labor-intensive, demanding both deep technical expertise and substantial time investment.
NEC’s system indexes the company’s internal patent information, technical documents, and past research reports using RAG. When a user submits a natural-language query, the generative AI automatically produces a candidate list of relevant patents, summaries, and preliminary assessment comments. Human experts are then able to focus on final judgment rather than information gathering, dramatically compressing turnaround time.
Why RAG Works for Patent Search
RAG (Retrieval-Augmented Generation) is a technique in which a large language model (LLM), when generating a response, retrieves documents from an external database and uses them as context. Unlike standard generative AI, which relies solely on its training data, RAG can incorporate the latest external information and proprietary internal data into its outputs.
RAG is particularly well-suited to patent search because patent documents are written in a highly specialized register with a distinctive structure. Patent claims, which carry legal force, are often misinterpreted by general-purpose language models. NEC developed optimized vectorization and indexing techniques for patent documents to improve both retrieval precision and generation accuracy. Equally important, the system supports on-premises and closed-cloud deployment to prevent leakage of trade secrets contained in internal documents — a critical requirement for enterprise customers.
A Structural Problem Driving Demand
The timing of NEC’s commercial launch is well-calibrated to market conditions. Japanese corporations face a severe shortage of qualified IP professionals: even large enterprises struggle to hire patent engineers and patent attorneys (benrishi). Meanwhile, patent filing volumes are rising, and competition for rights in complex technology domains — AI, semiconductors, biotechnology — is intensifying. In this environment, AI-driven automation of search tasks is shifting from a “nice-to-have” to a business necessity.
Globally, investment in IP AI tools is accelerating. Specialized patent search and analytics AI startups — including Annova IP and Unified Patents in the U.S. and QantIP in Europe — are proliferating. NEC’s commercial entry brings it into competition with these global players. The company’s strengths in processing Japanese-language patent documents and its domestic consulting infrastructure are genuine competitive advantages, but the depth of its coverage of international patent data (EPO, USPTO) will be a key determinant of success.
SaaS Plus Consulting: A Hybrid Delivery Model
NEC is not offering the service as a standalone SaaS tool but as an integrated offering combining AI technology with consulting services. The positioning reflects an understanding that IP transformation requires not only software deployment but organizational change management — guiding clients through the redesign of IP workflows themselves. The hybrid model, in which AI-generated outputs are refined by domain experts spanning prior-art search, prosecution strategy, and infringement risk assessment, is likely to resonate particularly with mid-sized corporations whose IP teams lack both headcount and specialized expertise.
NEC’s internal proof of concept — using the tool within its own IP department before commercializing it — provides a credibility advantage in customer conversations. The ¥300 billion FY2030 revenue target signals that NEC regards this as a core business line, not a peripheral experiment.
Open Questions at the AI-Patent System Intersection
The proliferation of IP AI tools raises unresolved questions that will shape the industry. One concerns the reliability of AI-identified prior art: if the system incorrectly flags a reference as relevant, downstream prosecution and business decisions may be compromised. A second concerns how AI-assisted searches are disclosed to patent offices — a point the USPTO has begun to address through examiner-guidance updates. A third is the long-term impact on the patent attorney profession. NEC’s service is framed as augmenting rather than replacing human expertise, but as AI capabilities advance, the division of labor between human specialists and AI systems in IP practice will continue to evolve.
コメント