Daiichi Sankyo is leveraging Azure OpenAI Service to develop generative AI and foster an AI-friendly culture

Please note that this interview was conducted on 26 March, 2024

Daiichi Sankyo Co., Ltd. is a global pharmaceutical company based in Japan that was established in 2005 through the business integration of Sankyo Co., Ltd. and Daiichi Pharmaceutical Co., Ltd. In order to achieve its purpose of “contributing to the enrichment of quality of life around the world,” Daiichi Sankyo has established its 2030 Vision of being an “Innovative Global Healthcare Company Contributing to the Sustainable Development of Society” as its ideal state, and is moving forward with the fifth 5-Year Business Plan to realize its transformation from a conventional drug discovery-oriented company to a global healthcare company. 

“In our fifth 5-Year Business Plan, we have established four strategic pillars, and we are making company-wide efforts to achieve our targets with these pillars as growth drivers,” says Mr. Yusuke Asoh, Associate Director of Company-Wide Transformation Promotion Group, Digital Transformation Planning Department, Daiichi Sankyo Co, Ltd. One of the Daiichi Sankyo’s key foundations is the “data-driven management through DX, and company-wide transformation through advanced digital technology,” and the Advanced Digital Technology Team led by Mr. Asoh is promoting the search for and introduction of advanced digital technologies from the perspective of promoting company-wide transformation.  

Daiichi Sankyo promoting DX with ABCDX as the central point of transformation 

“The DX that Daiichi Sankyo aims for is represented by ABCDX, which includes four transformations (Xs). AX represents transformation of daily business activities (Activity Transformation), BX represents transformation of the ways we drive business (Business Process Transformation), and CX represents transformation of company’s culture (Culture Transformation). By promoting these three Xs in an integrated manner, we will realize the transformation of Daiichi Sankyo (Daiichi Sankyo Transformation (DX)). For us, DX is both digital transformation and Daiichi Sankyo transformation," says Mr. Asoh.

Advanced Digital Technology Team has created an environment that enables secure use of image AI and translation AI technologies on a company-wide level and has also led efforts for early use of blockchain and quantum computing technologies. Daiichi Sankyo’s series of efforts to promote DX, including these advanced and proactive measures, were highly evaluated, and the company was selected as “DX Stock” in 2023. It can be said that the company has become a company that leads the industry also in terms of digital utilization. 

Launch of project to develop “DS-GAI,” the company’s own generative AI system 

Mr. Asoh and his colleagues began to pay attention to AI around 2015. Around that time, utilization of deep learning technology for business purposes was gaining increasing attention, and they saw great potential in the emergence of AI technology that exceeded human accuracy in specific areas. For Mr. Asoh, who had been following AI trends since that time, the impact of generative AI was enormous.  

“Traditional AI is purpose-specific, and while it was useful for sophistication of limited areas, it was not easy to utilize it at all. On the other hand, generative AI such as ChatGPT enables a world where anyone can use AI even without massive amounts of data or expertise. I felt in my gut that AI would be indispensable for companies to maintain and gain competitiveness in the future," said Mr. Asoh.

Within Daiichi Sankyo, momentum for utilizing generative AI for business purposes grew rapidly, and the company began to consider utilizing it. However, at the same time, the company was also faced with issues to be resolved, such as misleading and information leakage due to erroneous output from AI, infringement of copyrights of third parties, and misunderstanding of and overconfidence in generative AI. 

Therefore, Daiichi Sankyo first established guidelines for the use of online services including ChatGPT and made them known to the entire company. In addition, based on the company’s decision that its own generative AI platform is indispensable to fundamentally solve various issues related to generative AI, the company started developing “Daiichi Sankyo – Generative AI (DS-GAI),” an in-house generative AI system that utilizes Microsoft’s Azure OpenAI Service. 

Reliability of Azure OpenAI Service that dispels security concerns 

As a pharmaceutical company that often handles sensitive information, Daiichi Sankyo has a rooted culture of thorough information management. Therefore, security was one of the most important issues to be solved during the development of DS-GAI. “Even if we could technically achieve a high level of security, we thought that if the users themselves were not convinced that they could use this system with a sense of security, it would not be accepted within the company,” said Mr. Asoh. 

Valuable advanced technology is meaningless if it is not used. It was then that Azure OpenAI Service, a development platform, was selected as a candidate. It is a Microsoft product, one of the world-leading companies with past results of cloud platform management, and Mr. Asoh thought that users would accept Azure OpenAI Service with a sense of security because it allows systems to be built on a closed area network within a cloud tenant. 

“Azure OpenAI Service comes with a content filtering system and secure API authentication, allowing for secure deployment within the company. Additionally, Azure AI Search and other resources are available, making it attractive for its ability to flexibly and quickly respond to future function enhancements," said Mr. Asoh.

Selected Avanade as a partner to support system construction 

Daiichi Sankyo had little experience in using Azure as IaaS, so it did not have sufficient knowledge of building Azure infrastructure. Therefore, it selected Avanade Japan K.K., a Japanese subsidiary of Avanade, a joint venture between Microsoft and Accenture, as a partner for system construction because Avanade Japan has extensive experience in building systems on Azure. 

“We were looking for a company that had a proven track record, a strong relationship with Microsoft, and knowledge that we did not have. No one else could meet all these conditions but Avanade. Furthermore, another reason we selected Avanade was that they responded positively to our request to ‘release an in-house generative AI system in one month that could be used by all 9,300 employees of our company,’” said Mr. Asoh. 

It was May 2023 when Daiichi Sankyo created a long list for selection of a partner for development of its own generative AI system. By that time, ChatGPT had already begun to spread explosively around the world. Mr. Asoh says that there was a consensus among them including the management that they needed to quickly proceed with implementation of the system and that no delay was acceptable in order to gain a competitive advantage. Then, in early June 2023, Daiichi Sankyo approached Avanade about becoming a development partner, and the desired release date Daiichi Sankyo told Avanade at that time was September 1.  

Mr. Tatsuma Mihata, Director of Avanade Japan K.K., who led the development of this project, looks back at the development of the system. “To be honest, I have never heard of a project like this, which requires a company-wide release of a proprietary generative AI system for about 10,000 employees in just one month after the start of the project without a Proof of Concept (PoC) period.” 

Around that time, a generative AI boom was taking place in various companies. Mr. Katsuhiko Matsuno, Manager of Sales Division, Avanade Japan K.K., explains the background of the decision to accept Daiichi Sankyo’s request as follows. “Daiichi Sankyo’s project required extremely high-level skills and delivery capabilities, but we felt that if we could achieve this by concentrating our knowledge and technological capabilities including those from our global network, we would gain great confidence. So, we told Daiichi Sankyo we would love to accept their request.” 

“The risk was high, but we stood a fair chance of success. When Daiichi Sankyo approached us, we were already developing a common AI platform based on our past experiences. We had also developed our own framework, based on Microsoft’s Cloud Adoption Framework (CAF), that covered all the considerations when using Azure. We concluded that if we could keep development to a minimum by utilizing these assets, we would be able to meet the deadline, and we proposed this approach,” said Mr. Mihata.

Rapid development enabled production release for all employees in one month 

The DS-GAI project, which began as described above, was carried out in phases. Phase 1, which kicked off on August 1, 2023, first aimed for a production release for all employees on September 1. Thereafter, Phase 2 was a four-month project aimed at making improvements based on feedback from users and implementing more advanced functions. 

Looking back on Phase 1, Mr. Mihata of Avanade says, “The assets of the AI platform we had were basically for PoC, we faced some issues when we tried to bring DS-GAI closer to the production environment.” 

“The first issue was load balancing within the Azure infrastructure. How can we balance the load to cope with the constraints of Azure OpenAI Service, such as the number of tokens processed in a given period of time? The second issue was improvement in the management functions. The third issue was how to apply DS-GAI to the existing policies of Daiichi Sankyo’s authentication platform. In effect, we had to solve these issues in less than a month,” said Mr. Mihata.

The development team consulted with Microsoft’s staff in charge of architecture and internal experts of Avanade to solve the issues one by one and achieved the release of DS-GAI for all employees in Japan on September 1, one month after the project kicked off. 

Implemented high-order functions one after another through agile development, simultaneously proceeded with activities to make them known 

Phase 2 was carried out by agile development. “In the world of generative AI, updates are being made from hour to hour. We thought that traditional waterfall development methods would not be able to keep up with that evolution,” said Mr. Asoh.

A development system was set up to “release functions when they are ready,” and updates were made almost every month. When it was first released, the interface was simple for interacting with AI, but it was updated to “Daiichi Sankyo specifications” one month later in October based on feedback from users. Furthermore, in November, GPT-4, which had just been released, was implemented to improve the accuracy of DS-GAI all at once. In December, a function to upload and analyze internal documents was implemented. In January 2024, an image generation function was added to realize multimodal support, and in February, implementation of an internal data set analysis function and a Code Interpreter function was completed. DS-GAI was upgraded so that it can correctly respond to even events specific to Daiichi Sankyo and internally execute Python code, making it possible to be used in the science and technology field, which is the company’s strength. 

“Even though we requested it, I was surprised at the speed at which new functions were added rapidly every month,” says Ms. Junka Kaneda, Associate Director, Company-Wide Transformation Promotion Group, Digital Transformation Planning Department, Daiichi Sankyo Co, Ltd. 

Ms. Kaneda has been leading the dissemination of the functions of DS-GAI and support of its utilization in the DS-GAI project. Prior to the release of DS-GAI, Ms. Kaneda first held a lecture meeting in which experts talked about the significance of generative AI to attract the attention of employees. After the release, she held training sessions on how to use DS-GAI and workshops to create ideas for its utilization. Furthermore, in order to make the use of DS-GAI more applicable to business operations, she developed an in-house portal site, “DS-GAI Hub,” which allows users to view examples of advanced initiatives. 

“When the number of cases of pioneering utilization began to increase, we held in-house seminars and invited employees who were making full use of DS-GAI to speak and share their know-how. As a result, we began to receive more and more inquiries about, for example, how DS-GAI could be used in each department, and we are now actively holding training sessions for each department,” said Ms. Kaneda.

As a result of these dissemination activities, more than half of the Daiichi Sankyo Group’s employees have experience using DS-GAI at least once, and about 600 employees use it constantly every day. The main ways of using it include bouncing ideas off of each other, writing sentences, proofreading, and automatic generation of English sentences. According to a company-wide questionnaire survey, more than 80% of respondents said that DS-GAI has improved the productivity and accuracy of their work. 

“What makes me glad is that some of the employees say that ‘they have acquired new skills.’ Employees who do not know anything about programming are customizing applications, or employees who do not have artistic taste are generating illustrations. As these examples show, I believe that the introduction of DS-GAI has contributed in no small way to our ABCDX measures,” said Ms. Kaneda.

On the other hand, Ms. Kaneda says that there is room for improvement in the situation where some employees are not able to make full use of various functions of DS-GAI. As a result of a rapid succession of version upgrades due to agile development, there is a group of employees who are not able to fully catch up with the upgrades, and she thinks that approaching this group will be an issue in the future. 

“From now on, we would like to conduct thorough interviews with each department and propose ways to use DS-GAI that are appropriate for each department. I think that it is also necessary to analyze how to use new functions and models to be implemented in the future from the perspective of actual business operations, and to educate the employees about them,” said Ms. Kaneda.

DS-GAI has even higher expectations, PoC for Copilot service also progressing 

Daiichi Sankyo is currently in preparation for Phase 3 of the DS-GAI project. Mr. Asoh says that there is still much work to be done, as there are plans to continue to implement highly unique functions and improve the company-wide sophistication of the use of generative AI. 

“We would like to expand our platform so that we can use various Large Language Models (LLMs) according to the purpose of utilizing Azure AI Studio, and there is also room to improve the internal data set analysis function to make it easier to use. In order to promote the active utilization of generative AI in each specialized area and to realize the acceleration of DX across the company, we would also like to consider connecting DS-GAI to external databases and linking it with the internal business system,” said Mr. Asoh.

Mr. Asoh also expressed his high expectations for the future development of DS-GAI as follows: “We do not want to end with the release of DS-GAI, but to make it a driving force for company-wide DX promotion through repeated improvements.”

In addition to DS-GAI, Daiichi Sankyo is working on PoC to introduce Copilot for Microsoft 365. 

“I believe that the benefits of being able to seamlessly utilize generative AI within an integrated interface are enormous. Copilot Chat, in particular, can instantly generate new knowledge based on a wide range of information within a Microsoft 365 tenant, which I expect will contribute significantly to company-wide productivity enhancement,” said Mr. Asoh.

“I believe that generative AI will be involved in all aspects of our work and lives in the future. Rather than just being able to use Copilot, I would like to evaluate this trial with an emphasis on fostering skills and mindsets to make full use of generative AI naturally." said Ms. Junka Kaneda: Associate Director, Company-Wide Transformation Promotion Group, Digital Transformation Planning Department, Daiichi Sankyo Co, Ltd. 

Looking ahead to the future, laying the foundations for AI-ready state 

“I believe that in a few years, we will be in a world where the use of generative AI will be commonplace. That’s why we need to make multifaceted efforts to be able to respond to any situation and create a state in which we are ready for generative AI,” said Mr. Asoh.

Ms. Kaneda nodded in agreement and expressed her expectations for generative AI. “I think generative AI will be utilized not only for applications but also for a variety of things, such as infrastructure and middleware. I strongly expect Microsoft to expand such solutions.” 

Mr. Asoh also expressed his expectations for further evolution of Microsoft’s products as follows. “I feel that Microsoft has provided more than we expected. As the world-leading cloud AI developer capable of meeting strict requirements of the enterprise level, I hope that Microsoft will continue to expand functions and provide us with their support.”

Mr. Mihata of Avanade added reassuring words, saying, “I don’t want Microsoft to lose to other public cloud services. We will be responsible for supporting projects of our customers who choose Azure.” 

Daiichi Sankyo has positioned the promotion of DX as the foundation for its growth as a company and continues to pursue best solutions for itself while always looking ahead to the future. Microsoft Japan will continue to provide best support to and have a strong partnership with those who support the global healthcare industry in cooperation with its partners.