Developed an AI system for Cash Application in Azure to automate about 20% of manual work! The future of operations DX that the company works on

Initiated to enhance the efficiency of “Cash Application” as a project to utilize AI developed by the core system renewal project

FUJIFILM Business Innovation is expanding its business with its business philosophy of “Continuously innovate business ahead of change,” which is in line with the FUJIFILM Group’s corporate philosophy. Regarding the digital transformation (“DX”) initiatives undertaken by the entire group, there are three pillars: “Product/Service DX” centered on the shift from the MFP business to the solutions business, “Operations DX” to transform back-office operations, and “Human Resource DX” to develop human resources to facilitate the DX. Particular emphasis has been placed on operations DX, which supports product and service DX, and a DX Initiative Department has been established under the core system renewal project responsible for internal business reforms. The department has committed to various operations DX projects. Shigenobu Inenaga, Director and Senior Vice President of FUJIFILM Business Innovation, responsible for the core system renewal project, explains the role of operations DX that the company has promoted.

“The FUJIFILM Group covers a remarkably broad range of businesses, including the business innovation field that we operate, as well as medical, healthcare, imaging, and more. Each business has its own DX direction, but the FUJIFILM group as a whole has a shared policy of proceeding based on the three pillars of Product/Service DX, which revolutionizes the business itself, and Operations DX and Human Resource DX, which support the execution of the Product/Service DX. The same applies to Fujifilm Business Innovation. We regard product and service DX, which aims to transform the business itself, as our primary focus, and have launched a core system renewal project to promote operations DX to serve as the basis for this transformation. The project initiatives include implementing Dynamics 365 (Dynamics 365 for Finance and Operations, Dynamics 365 Customer Engagement) as the core system, and deploying Microsoft Azure (“Azure”) as main cloud computing system. We are now working to digitalize the entire business and improve its efficiency by connecting the core system and the current business with tools such as Microsoft Power Platform (“Power Platform”) and Microsoft 365,” said Inenaga.

Inenaga had believed that the improvement of business efficiency and automation through AI was a mission of extremely high priority in promoting operations DX in the core system renewal project. He recalled that a company-wide interview had been conducted in order to identify operations where AI could potentially solve their issues. Then, more than 40 tasks were picked up, and their priorities were examined by estimating the effects of AI utilization. One of those tasks was “Cash Application.” Toshihiko Fushimi from the DX Initiative Department, who joined the core system renewal project and has been involved in the promotion of operations DX, describes the story of the project.

“We conducted interviews with various departments, including sales, planning, procurement, after-sales service, and affiliated companies, about their business issues, and completed around 10 PoC projects, including spare parts supply and demand forecast, voice data text conversion, and many more. Then, we selected Cash Application as one of the most essential operations. As Cash Application had required a heavy manual workload and been partially outsourced, we concluded that the AI utilization would provide a significant return on investment,” said Fushimi.

The Cash Application of accounts receivable by linking payment information to invoices requires a very time-consuming and labor-intensive process. Particularly at FUJIFILM Business Innovation, which handles more than 100 thousand receipts per month and as many as 1.2 million invoices per month, the workload demands tremendous amount of efforts. Shin Sakaida from the Transformation Promotion Group, Service Planning Department, FUJIFILM Service Creative, who is in charge of the company’s contact center operations and invoicing, said, “The monthly process of linking 100 thousand cases of payment information to invoices requires considerable time and efforts, and during the busy end-of-month and beginning-of-month periods, the dedicated personnel alone could not handle the entire process. Consequently, we had no choice but to assign the work to other teams. We had to rely extensively on the experience and intuition of our senior staff to locate customer accounts and check combinations of amounts,” explaining the high workload involved in the Cash Application procedures.

Having been involved in accounting-related work for many years, Inenaga was well aware of ‘how tough’ the task of Cash Application was. He therefore hoped that the AI utilization would solve this problem, recalling when the project was initiated.

“Owing to the enormous number of customers and the complexity of the business, Cash Application at FUJIFILM Business Innovation has become an extremely demanding task. Some customers invoice multiple contracts at once, which often makes it difficult for us to handle contracts, invoicing, and payments. The process is akin to linking 10 invoices out of 1.2 million invoices for each payment. Naturally, we have introduced tools to improve efficiency, but there remain areas that require more than that. Hence, we considered this to be an effective AI utilization project,” said Inenaga. 

Adopted Azure as the underlying platform for the AI system and smoothly promoted the project through parallel development by the technical and operational departments

Once adopted as part of the Operations DX, the Cash Application AI project started in the summer of 2020. Since Dynamics 365 had already been chosen as the core system at an early stage of the project, Azure was the most likely candidate for the cloud service that would serve as the system platform.

“We considered revamping our core system when its maintenance expired, and we chose Dynamics 365 for its cloud native features and excellent integration with BI tools and Microsoft Office applications. Therefore, it was inevitable that the cloud service forming the underlying platform for the AI system would also be Azure. Although we compared Azure with other cloud services, we eventually decided to opt for Azure as the best option considering its overall performance and the availability of excellent tools such as Power Platform,” said Inenaga.

To build the system, we developed our own AI system instead of using Azure’s AI service, due to the need for customization, such as incorporating business logic and applying it to the in-house security policy. The system was finally built by using VMs on Azure, Azure Storage, Azure Machine Learning, and Azure CLI. Masato Ando from the Technology Development Group, Business Solution Service Division, who had driven the technology development for AI utilization for the core system renewal project and was in charge of AI technology development and system development for the Cash Application AI project this time, explained the system configuration for the Cash Application AI as follows.

“We configured the system so that Azure and Dynamics 365, the core system, would work together with RPA. Since no mistake is allowed in accounting projects, we devised a method of dividing the processes to be automated and those to be done manually in accordance with the degree of AI’s confidence. For this part of the project, the operational department builds with Power Automate, while the AI part is built by the technical department, thus accelerating the project speed,” said Ando.

Sakaida has classified the processes to be automated and those to be done manually from the perspective of accounting personnel workers who actually perform the Cash Application work. He described the challenges of being involved in the project as an operational department member, saying that it was necessary to change the perception of AI.

“Due partially to my lack of knowledge, I assumed that the answers given by the AI were 100% correct, and struggled to deal with the fact that some of the answers were wrong. Considering operational use, we must avoid even a single mistake. Therefore, we incorporated the business logic piled up over the years and built a mechanism to extract only correct answers from a mixture of correct and incorrect ones. The results generated by the AI are then re-calculated using logic so that the processing can be divided between automatic and manual processing,” said Sakaida.

Ando said that such establishment by an operational department has led to a synergistic effect in the development of AI as well. He added “Through interviews with the field, we identified features that were directly related to accuracy. We also repeatedly discussed what the necessary parameters were for areas where the results appeared unreasonable, and proceeded to improve the AI in an agile development cycle,” attributing the effectiveness of the project to close communication between the operational and technical departments. In fact, it is not an easy task to determine the amount of features that had become tacit knowledge of senior professionals through interviews, but the company has been engaged in AI research and development for over several decades, and its experience and know-how have been utilized to smoothly promote the project. In addition, Ando noted that Microsoft has provided intensive supports for the project, including the development of their own AI, and he was pleased to receive Microsoft’s assistance in a variety of areas, such as scaling between busy periods and the rest of the year, and the mechanism for maximizing Azure’s performance with assurance of security.

Accelerating approaches to data utilization with Microsoft products while promoting transformation of the business itself

Supported by Microsoft, the Cash Application AI project, facilitated through collaboration between the technical and operational departments, has progressed smoothly and has entered into official operation in February 2022. Previously, 66% of the Cash Application work had been done automatically with the use of tools and logic, but this project has improved the efficiency for the remaining 34% of the manual Cash Application work. The AI performed tasks previously relied on the experience and intuition of senior staff, leading to a significant reduction in the number of manual operations. At the PoC stage, we had targeted to automate about 10% of the manual Cash Application work, but through continuous improvements by both the technical and operational departments, we have now succeeded in automating about 20% of the work, significantly exceeding the target. Sakaida explained the results of this project, saying that accounting personnel were very pleased with the change. Ando also noted the benefits of introducing AI, saying, “Even beyond the areas we have automated, we have achieved over 70% correct answers in generating candidates for Cash Application, reducing the amount of time spent on manual processing,” adding that the company has acquired core AI technology to improve the efficiency of back-office operations through this initiative.

Fushimi, who has promoted operations DX, said, “Although we have already applied AI in business departments, such as in the MFP development, the project was the first time that we achieved results with AI utilization in the operational department.” He hoped that this project serves as a case study of how AI utilization can lead to further advanced BX (Business Transformation) deployment.

For the future prospects, Fushimi expected to accelerate AI utilization with Microsoft products such as Azure and Power Platform, and to reformulate the company’s internal operational procedures. More specifically, the company plans to release a system using Azure’s voice recognition service so that transcription of voice data, which was previously outsourced to contractors, will be performed in-house and more efficiently, he said.

Ando also mentioned that, “We have introduced Dynamics 365 as our core system, and we also accelerate the use of Azure. Therefore, we would like to continue to leverage services provided by Microsoft not only in the technical department but also accounting personnel,” and cited the Azure OpenAI Service, which provides access to OpenAI’s GPT-3 model, as an Azure service that he was personally keen on.

Inenaga, supervising the core system renewal project for operations DX at FUJIFILM Business Innovation, described the DX vision to which the company looks ahead, as “the transformation of the business itself” and then talked about the future prospects.

“In the past, our business mainly entailed MFP, but as I mentioned at the beginning, we need to change our business itself under current circumstances where digitalization is progressing and paper-based communication has gradually declined. I believe that it is an excellent opportunity for us to change our business structure at a time when the DX trend has arrived. The principal focus is on solutions to deliver DX for our customers. For example, when selling Dynamics 365 externally, we believe that the goal of product and service DX is to comprehensively support customers, including integration and the direction of business improvement that is the initial step for such process. On the other hand, the key mission in operations DX is to create an environment for everyone to utilize all business data in a centralized manner. The centerpiece of the mission is the core system, Dynamics 365, and we are now in the process of its implementation and the development of related systems to make better use of data. We assume that AI has an important role to play here, and that the leveraging of Microsoft products including Power Platform and Microsoft 365 will be critical,” said Inenaga.

We will have to keep a close eye on FUJIFILM Business Innovation’s efforts to promote product/service DX, operations DX, and human resource DX with Dynamics 365 and Azure services adopted for the core system.