Ricoh is transforming from a multifunction printer (MFP) manufacturer into a digital services powerhouse with a strong focus on artificial intelligence (AI). At the center of this shift is its proprietary 70-billion parameter large language model (LLM) and an investment in training employees to deploy AI-driven solutions for customers. Ricoh has prioritized AI innovation in Japan, rolling out solutions ahead of its European and U.S. markets.
Building AI Talent: The AI Evangelist Certification
In 2024, Ricoh Japan launched an AI Evangelist certification system to foster AI expertise among its workforce. Of the 1,387 employees pursuing certification, only 300 are expected to meet the rigorous qualification by 2025. Ricoh plans a Senior AI Evangelist certification for advanced practitioners in 2026. This initiative aligns with Ricoh Japan’s broader reskilling efforts to support the growing demand for AI solutions.
Decades of AI Expertise
Ricoh’s AI development dates back to the 1990s, with deep learning advancements since 2015. Its early AI applications used image recognition for quality control and vibration monitoring. Since 2020, Ricoh has leveraged natural language processing (NLP) to analyze office documents and customer feedback at call centers. By 2022, the company intensified its focus on LLMs, unveiling its proprietary model in 2023. The latest iteration, a 70-billion parameter LLM, supports Japanese, English, and Chinese and offers cloud-based and secure on-premise implementation. The company is also advancing research in voice recognition AI and offering customers AI agents with voice interaction capabilities.
Key Features of Ricoh’s 70-Billion Parameter LLM
- Optimized for Japanese, with Multilingual Capabilities. Ricoh’s LLM is trained using proprietary corpus selection, data cleansing, and curriculum learning to ensure precise Japanese responses. Co-developed learning scripts with AWS enhance its English and Chinese capabilities, which allows it to understand diverse Japanese, English, and Chinese expressions. Further training with approximately 16,000 instruction-tuning data points, including proprietary developments, has enhanced its adaptability to various tasks. To ensure high-quality private LLM development, this approach minimizes performance degradation due to catastrophic forgetting during additional learning when building private LLMs tailored to customer needs.
- Efficiency Gains Through Advanced Tokenization.* A customized tokenizer improves Japanese language processing by 43%. The tokenizer divides text into tokens for LLM processing, enhancing efficiency and reducing resource usage, shorter response times, and cost savings. Given the significant power consumption and environmental impact of LLM processing, this technology also contributes to energy conservation and reduced environmental load.
- Secure On-premise Deployment. Typically, operating and training a 70-billion parameter LLM requires a large cluster system connecting multiple servers via a network. Ricoh’s LLM leverages unique vocabulary replacement and other cutting-edge technologies to enable training while maintaining model size. For customers prioritizing security and data retention within their premises, additional training, including confidential information, is possible in closed environments
- Cost and Energy Savings. With support from the AWS LLM Development Support Program and the AWS Generative AI Innovation Center, Ricoh developed its model using Amazon Elastic Compute Cloud Trn1 instances equipped with AWS Trainium accelerators. This enables more affordable and timely delivery of custom LLMs to customers. Additionally, using Trn1 instances for training improves energy efficiency by up to 25% compared to equivalent accelerated computing EC2 instances.
Training an AI-savvy Workforce
With 18,000 employees across 48 branches and 358 sales offices in Japan, Ricoh Japan has a strong foundation for AI deployment. The workforce includes about 7,400 sales staff, 4,400 customer engineers, 1,300 system engineers, and 1,000 contact center personnel. This structure allows staff to visit customers nationwide, identify their challenges, and propose improvements. In 2024, the company introduced an internal specialist certification system across five IT fields, including AI Evangelists. Unlike previous ICT (Information and Communications Technology) roles, the new system encourages employees to develop expertise through real-world applications.
AI Evangelists gain certification by completing external qualifications like the G Test, Ricoh Japan’s e-learning content, and demonstrating AI knowledge through internal and external seminars, solution sales, and business improvement initiatives. Candidates use Ricoh’s internal data utilization services, RICOH Chatbot Service Digital Buddy** and RICOH Digital Buddy using RAG (Retrieval-Augmented Generation) technology and no-code generative AI app development platforms to engage in AI operation and development, create use cases, and drive internal business improvements. These specialists can make advanced AI proposals to customers by acquiring more practical knowledge and skills.
*A tokenizer is a tool used in NLP to break down text into smaller units called tokens, by adding features or optimizing algorithms to handle specific language complexities or improve the accuracy and efficiency of the tokenization process.
**RICOH Chatbot Service Digital Buddy, a service provided by Ricoh, leverages generative AI to utilize internal data like a trusted partner.
