Efficiency, quality improvement, and cost reduction achieved through data annotation outsourcing ―― Improved efficiency of AI development through division of labor, contributing to shortened development period.

Interview Cooperation: Mr. Keita Kawakami, Business Design Group, Head of Business Development Department

SCSK Corporation Overview

・Established: October 25, 1969

・Capital: 21,285 million yen

・Number of Employees: 15,328 (consolidated), 8,470 (non-consolidated) (as of March 31, 2023)

・Business Content: SCSK Co., Ltd. provides all necessary IT services for business, including consulting, system development, verification services, IT infrastructure construction, IT management, IT hardware and software sales, and BPO (Business Process Outsourcing), with a full lineup.

Services

・Data Annotation Service

Challenges before implementation

HS: Please tell us about the project and commission content that you have requested from our company.

Mr. Kawakami: This project aims to demonstrate the use of AI for anomaly detection in order to improve the efficiency of maintenance operations being carried out manually in a certain industry.

The three points of the commission are as follows:
・Data annotation (rectangular/semantic segmentation)
・Pickup (selecting images with the subject in them)
・Generating difficult-to-collect abnormal images (using image generation AI, image editing software, etc. to create composite images)

HS: If you have any issues that you feel while working on the project, please let us know.

Mr. Kawakami: Yes, there were several issues. First, last year, half of the data annotation work was done in-house, and the remaining half was outsourced to external companies.

Increase in Time Spent on Data Annotation
One of the reasons is the shortage of manpower. Data annotation requires a lot of effort and time, making it difficult for our team to handle the workload with our current number of members.

By improving the quality of data annotation, we aim to improve AI accuracy.
The second issue is that our previous system did not have specialized data annotators, making it difficult to establish and improve annotation standards that would yield consistent results regardless of who performed the annotation. This resulted in variations in the quality of annotation outputs. The low quality of annotation was one of the reasons for the lack of improvement in AI accuracy.

Reasons for Choosing Human Science

HS: Can you tell us about the reason for requesting HS?

Mr. Kawakami: Yes. In order to address the previous challenges, we have started considering outsourcing to a knowledgeable data annotation company. However, at that time, there were no companies registered as our outsourcing partners for data annotation work.
Therefore, we first started by researching companies that perform data annotation. We initially selected about 60 companies and requested materials from them. At this stage, we narrowed down to about 40 companies based on factors such as financial status, credibility, industry share, future prospects, and risks. We then set 20 evaluation criteria and assigned points, ultimately narrowing down to 10 final candidates.

Afterwards, we directly contacted these selected companies and conducted detailed hearings and data annotation trials. As a result, we narrowed it down to three companies and ultimately chose your company.

HS: Please tell us the biggest reason for choosing our company.

Mr. Kawakami: One of the most important factors was finding a company that we could build a long-term business relationship with. Not only for this project, but also for the possibility of outsourcing data annotation work from various departments in the future, it was necessary to select a company that could handle multiple and large-scale projects.

By entrusting your company, we were able to solve issues such as quality that we had been facing, and while working together, we were able to absorb that knowledge and ultimately contribute to the improvement of AI accuracy, which we had hoped for.

Introduction Effect and Evaluation of Human Science

HS: Did you see any results after implementation?

Mr. Kawakami: Although it may reveal our expectations for your company, here are some of the benefits we have experienced since requesting your services.

＜1. Focusing on AI development＞
When we were doing annotation work in-house, it took a lot of time. It was difficult to establish annotation standards, and there were many revisions and unnecessary work. By outsourcing the annotation work, we were able to create an environment where we can focus on AI development. This allowed us to concentrate our resources on development and improve efficiency. Being freed from annotation work allowed us to prioritize development.

＜2. Higher Quality Data Annotation Results Obtained Than In-house＞
We were expecting high-quality results for data annotation, which was difficult to achieve in-house. By using the criteria prepared here and consolidating Q&A and feedback between your company and ours during the progress of the work, the criteria became even clearer. As a result, the quality of data annotation has improved compared to in-house work, greatly contributing to the improvement of AI accuracy.

＜3. Shortening Development Time through Division of Labor＞
Due to the limited number of engineers that can be secured in-house, it was a challenging task to simultaneously conduct data annotation and AI development without any room for development time. However, by outsourcing data annotation to your company and dividing the work, we were able to focus on AI model development and accuracy improvement, and successfully complete AI projects in a short period of time.

＜4. Acquisition of Knowledge on Data Annotation Work＞
We have gained a lot from your company in terms of sharing knowledge on data annotation. Through our meetings, you have taught us specific methods for data annotation, perspectives to consider when performing the work, and the tools to use. In particular, this time, you gave us a demo and lecture on specific methods to improve segmentation and work efficiency, which has contributed to our knowledge improvement. We also highly appreciate your company in these aspects.

<5. Cost Reduction>
Finally, we were able to benefit from cost savings and contribute to the profitability of the project. Overall, we evaluate that outsourcing the annotation work had a positive impact in many aspects.

HS: Please tell us about your experience with our company, from the time of implementation to the present.

Mr. Kawakami: To be honest, I don't have many challenges or concerns with your company, and I am currently satisfied with everything. I have a very good impression. As for data annotation, I would like to continue to ask your company as much as possible within the budget.