Sales Retriever Achieves 6x Performance of Latest OpenAI Models in Enterprise Key Person Research with Proprietary Patent
Sales Retriever has announced that it achieved 6x the performance of the latest OpenAI models in enterprise-focused key person research using its patented technology (JP 2026-52570).
📋 Article Processing Timeline
- 📰 Published: May 25, 2026 at 19:10
- 🔍 Collected: May 25, 2026 at 10:31
- 🤖 AI Analyzed: May 25, 2026 at 22:39 (12h 8m after Collected)
Sales Retriever, provider of AI for enterprise business development (Headquarters: Chiyoda-ku, Tokyo; Representative Director: Nariyuki Matsumoto; hereafter "Sales Retriever"), announced that it has achieved 6x the performance of the latest OpenAI models in enterprise-specialized key person research using its patented technology (JP 2026-52570).
■ Background of the Patent
In sales activities targeting large enterprises, accurately grasping the target company's organizational structure and identifying the appropriate departments and key persons is crucial. However, large enterprises have complex organizations including group companies, subsidiaries, and affiliates, and department names vary, making it difficult to maintain accurate contact information.
While generative AI and web search-based corporate research are spreading, general AI research often faces issues such as mixing information from similar companies, insufficient evidence sources, or decreased accuracy when surveying multiple companies simultaneously.
Sales Retriever developed a department research technology to solve these issues by evaluating target companies, department names, positions, names, and evidence sources combined to extract actionable contact information.
■ Verification Results Compared to Latest AI Models
In this verification, five departments (HR, IT, General Affairs, Sales, and Production Planning) were surveyed across scales of 1, 5, 10, and 20 companies. Sales Retriever was compared against representative models including GPT-5.5, Gemini 3.5 Flash, and Claude Opus 4.6 in terms of "number of correctly extracted contacts," "accuracy rate," and "execution time."
Sales Retriever demonstrated superiority over general AI models across all metrics. In a 20-company simultaneous search, it achieved 6.1 times the number of correct contacts compared to GPT-5.5, the best-performing general model.
While other models showed significant performance degradation when searching multiple companies, Sales Retriever maintained a high accuracy rate of over 90% in all scenarios. Regarding execution time, it consistently completed searches in approximately 60 seconds, which is 1/5 of the time taken by GPT-5.5.
■ Background of the Patent
In sales activities targeting large enterprises, accurately grasping the target company's organizational structure and identifying the appropriate departments and key persons is crucial. However, large enterprises have complex organizations including group companies, subsidiaries, and affiliates, and department names vary, making it difficult to maintain accurate contact information.
While generative AI and web search-based corporate research are spreading, general AI research often faces issues such as mixing information from similar companies, insufficient evidence sources, or decreased accuracy when surveying multiple companies simultaneously.
Sales Retriever developed a department research technology to solve these issues by evaluating target companies, department names, positions, names, and evidence sources combined to extract actionable contact information.
■ Verification Results Compared to Latest AI Models
In this verification, five departments (HR, IT, General Affairs, Sales, and Production Planning) were surveyed across scales of 1, 5, 10, and 20 companies. Sales Retriever was compared against representative models including GPT-5.5, Gemini 3.5 Flash, and Claude Opus 4.6 in terms of "number of correctly extracted contacts," "accuracy rate," and "execution time."
Sales Retriever demonstrated superiority over general AI models across all metrics. In a 20-company simultaneous search, it achieved 6.1 times the number of correct contacts compared to GPT-5.5, the best-performing general model.
While other models showed significant performance degradation when searching multiple companies, Sales Retriever maintained a high accuracy rate of over 90% in all scenarios. Regarding execution time, it consistently completed searches in approximately 60 seconds, which is 1/5 of the time taken by GPT-5.5.
FAQ
Sales Retrieverのキーパーソンリサーチ技術の特長は何ですか？
対象企業、部署、役職、人物名、根拠ソースを組み合わせて評価することで、別会社情報の混入や根拠不足を抑制し、高い精度と網羅性を実現しています。
一般的なAIモデルと比較した検証結果は？
20社同時リサーチにおいて、GPT5.5と比較して6.1倍の正答数を記録し、実行時間も1/5に短縮、回答精度は90%以上を維持しました。
なぜ複数社同時リサーチで他モデルの性能が劣化するのですか？
一般的なモデルでは企業ごとの正確な判定が難しく、引用不足や別会社との混同が発生しやすいためです。
この技術が活用される場面は？
大手企業開拓において、複雑な組織構造を持つ企業群から正確な担当者情報を抽出する営業活動の効率化に貢献します。
特許番号は公開されていますか？
はい、特許技術として「特開2026-52570」が取得されています。