Lessons from the AI Helpdesk Implementation and Applying Them to BPR | Insights

Generative AI has attracted a lot of attention since its first appearance, but implementation on the company and organizational level is still in the trial and error stages. ABeam Consulting is also receiving many inquiries about generative AI implementations, but there are few confirmed cases of significant improvements in work efficiency following its introduction. Best practices are still being worked out at many companies. Under these conditions, we introduced an AI Helpdesk service for accounting work at our company and greatly improved efficiency in those operations. In this article, we will share key points for generative AI utilization and our future outlook on this technology, based on what we learned by introducing the AI Helpdesk service.

(Restructured based on the following article released in June 2025 by PKSHA Technology Inc.: “Automating 90% of Inquiries from 7,500 People with an AI Helpdesk: Achieving Customer AIBPR Through Reforms at Our Company.”)

About the Author

Chihiro Nishioka

Principal
Kiichiro Kumada

Senior Manager
Shim　Hajae

Senior Consultant
Tetsuya Yoshimura

Consultant

1. AI Utilization Issues in Companies

Population decreases and worker outflows in Japan are exacerbating hiring difficulties and labor shortages, and there is an urgent need to boost productivity per worker. Workers are limited, but the amount of information and procedures continues to increase, and handling internal inquiries across departments such as accounting, personnel, and sales support is becoming more onerous, making operations more personalized and inefficient. There are high expectations that generative AI will solve these issues, but implementation is still in the trial and error stage. Some of the main concerns raised regarding generative AI in the questionnaire for attendees of the PKSHA Technology Inc. July 2025 webinar were “lack of skills within the company” and “security and information leaks,” and many companies have not yet implemented the technology. Our clients have also stated, “we made it to the PoC stage, but were unable to show effectiveness for work operations, so it’s stalled.” For example, one company implemented generative AI to improve work efficiency, but usage is limited to company-wide generic use cases, preventing PoC for department-specific expectations such as streamlining and enhancing operations. Another company expected to apply generative AI to use cases in the Sales Department, but insufficient knowledge of AI-specific system design and data management left the project stuck in the planning phase.
To resolve such on-site problems, it is essential to have a framework for considering user value while circulating knowledge. Our approach, as demonstrated in the accounting AI Helpdesk, could serve as a concrete measure for achieving this. The next section covers the background and process of implementing the system.

2. Background of the AI Helpdesk Implementation

As stated in the previous section, many companies in Japan are struggling with labor shortages and complex work operations. Our internal inquiry handling operations are no exception, with questions arising from accounting, personnel, and sales support flooding supervisors. This issue is exacerbated by lack of personnel and complex work operations. Since the Accounting Department was receiving approximately 100 questions per month about expenses calculations and only had 3-4 staff members responding to these emails on top of their regular work, slow and duplicated responses had become commonplace.
The content of these emails was not the type of structured data which generative AI could handle easily, and even considering FAQ site and although the company considered Retrieval-Augmented Generation (RAG) board maintenance, there were simply not enough labor hours available for data structuring.
To resolve this problem, we decided on a policy to organize and aggregate the inquiry content and responses, and to convert them to a data-driven framework that could reuse this content. The goals were to maintain service quality with limited personnel and to utilize knowledge as an organizational asset.
When selecting a solution, we did not stop at a simple knowledge search. We focused on achieving response quality above RAG levels at a low operating cost by combining AI and human responses, while automatically structuring inquiry logs as learning data. The system which met these requirements was the “PKSHA AI Helpdesk” (hereinafter: AI Helpdesk), a knowledge management service with integrated generative AI.
The AI Helpdesk combines three functions: 1) receiving questions through chat messages, 2) searching the company FAQ and related documents to present an answer, and 3) automatically aggregating conversation logs and proposing FAQ additions and revisions. Since human response escalation is built in as a standard feature, inquiries requiring individual handling can also be resolved through the same channels. The ability to implement the system over Microsoft Teams and use the existing communication environment without any changes was another reason we selected it. The next section considers the quantitative and qualitative results, sharing the specifics of what the system achieved.

3. Impact of the AI Helpdesk Implementation

The results of the AI Helpdesk implementation can be broken up into two categories. The first was streamlined operations. The Teams-driven chatbot was established as a first-line response through its FAQ and document searching capabilities. Inquiries increased by about 220% prior to implementation, but the number requiring human responses was reduced to just under 40% of the former numbers. Even with the former 3-4 member team reduced to an Accounting Department supervisor, they were still able to handle responses. Lead time for late night and holiday inquiries was dramatically shortened. The increase in inquiries was absorbed by the AI, and maintaining service levels while reducing human handling to a minimum provided a fixed quantity of learning data to back up the scalability and processing function of the AI Helpdesk.
The second was lowering barriers to usage and revitalizing the inquiry system. Shifting the inquiry channel from emails to Teams integration allowed employees to “casually ask questions and get an immediate answer.” This lowering of psychological barriers led to an increase in the number of inquiries, creating a workflow in which only about 20% of inquiries could not be handled by AI were escalated to the supervisor. Employees praised the rapid response of the chat system, and access for inquiries late at night and on holidays, which they would normally have to wait until the next business day, increased.
The combination of the weekly conversation log analysis and FAQ improvement proposal function created a continuous improvement cycle, which supported these results. By regularly reviewing logs, the system organized variability in notation and repeated questions, incorporating a reply text refinement procedure in its operations. Automatic reply rates have remained at a high level since the system was introduced, and quality has improved. Since the chatbot’s answers are updated based on accumulated data, users get the latest information each time they access it, eliminating complaints like “the information is old” and “I can’t find the answer.”.
Streamlining operations and lowering barriers to usage suggests that the AI Helpdesk is not just a FAQ automation tool; it functions as a platform that promotes knowledge circulation. Even though the number of inquiries increased, there was no strain on processing. The user experience improved, an excellent result for horizontal expansion within the company and for proposals to clients. The next section covers our future outlook for company-wide expansion and client company BPR support based on these results.

4. Future Prospects for Client AI x BPR

“Streamlining operations” and “lowering barriers to usage” stated in the previous section suggest that the AI Helpdesk could optimize all work within the company, not just in the Accounting Department. We are currently structuring inquiry data for General Affairs Department and intend to eventually utilize the AI Helpdesk system through bots on shared Teams channels. Organizing and linking Q&A data for each department by purpose will expand the knowledge base available for the generative AI to reference, which will improve response quality and expand the utilization scope to derived functions such as searching documents and proposing workflows.
AI Helpdesk Desk system is not a tool that remains stagnant after introduction. It is a solution that continuously accumulates knowledge used by employees in day-to-day operations and creates value by improving answer accuracy using this data. If the notion of “ask the AI Helpdesk if you’re having trouble” becomes an established action pattern company-wide, the conversation log volume and quality will continue to improve the system’s accuracy and operational processes. Continuously organizing and announcing internal verification results and improvement processes will create a solid backing for proposing the system to clients.
Our plan for future AI service deployment is set up in three stages, based on the foundation of established cases of internal implementation. In the first stage, we implemented PoC in the Accounting Department and confirmed the results of primary response automation and knowledge aggregation. Through this initiative, we built an internal track record for streamlining operations through AI utilization and accumulated expertise for considering its implementation. In the second stage, we will conduct results verification inside and outside the company, applying the gained expertise to adjacent departments such as General Affairs and introducing the AI Helpdesk system to major clients. We will strengthen our activities to propose optimized solutions for client needs by expanding our internal use cases. In the third stage, with the AI Helpdesk Teams chat function as a starting point, we will introduce generative AI and AI agents for successive operations, promoting AI x BPR outside of inquiry handling, and expanding to AI-driven client business reform support.

Generative AI is evolving rapidly, with new services appearing every day. It’s highly likely that functions that do not exist today will be in use at companies six months from now. We will keep up with these changes, supporting gradual automation and standardization at organizations struggling with labor shortages and complex work operations. ABeam Consulting combines digital technology and innovation, swiftly resolving social and management problems as a creative partner to our clients.

Insights TOP