SCSK Corporation and Insight Edge Co-Develop AI-Powered Skill Evaluation System for Specialty Certification Program
SCSK Corporation has jointly developed an AI-powered skill evaluation system with Insight Edge Inc. for its unique 'Specialty Certification Program.' This system leverages generative AI to replace the expert-led evaluation process, aiming to significantly reduce workload and enable timely, fair skill visualization.
📋 Article Processing Timeline
- 📰 Published: May 8, 2026 at 21:00
- 🔍 Collected: May 8, 2026 at 12:32
- 🤖 AI Analyzed: May 8, 2026 at 12:51 (19 min after Collected)
SCSK Corporation (Headquarters: Koto-ku, Tokyo; Representative Director and Executive Officer, President: Takaaki Toma; hereinafter SCSK) operates its unique 'Specialty Certification Program,' where internal and external experts evaluate and certify employees' specialized skills from the perspectives of fundamental abilities, knowledge, and practical experience. This time, Insight Edge Inc. (Headquarters: Chiyoda-ku, Tokyo; Representative Director CEO: Junichi Kosaka; hereinafter Insight Edge), in collaboration with SCSK, has developed a 'Skill Evaluation System' that utilizes generative AI to replace this expert-led evaluation process (patent pending by SCSK: Patent Application No. 2026-040744).
This system ensures the validity and consistency of evaluations and achieves fair and flexible skill assessment through multiple AI verification processes and evaluation logic based on accumulated skill definitions. By introducing this system, SCSK will significantly reduce the man-hours required for evaluation, immediately visualize employee skills acquired through work experience, and contribute to further promoting human capital management by utilizing these insights for timely job assignments that foster employee growth and maximize their abilities.
Background and Purpose
SCSK has implemented the 'Specialty Certification Program' as a mechanism to support the continuous growth and career development of each employee by visualizing their specialized abilities across 18 job categories, 41 fields, and 7 levels.
Annually, skill levels are comprehensively assessed through application documents detailing employees' owned skills and practical experience, and interviews with applicants conducted by experts in each field acting as examiners. Detailed level definition documents, which specify the necessary skills for each job category and field, serve as long-term career development indicators for employees, and feedback from examination results is utilized by employees as a guide for self-improvement.
On the other hand, challenges with this examination process, involving application documents and expert interviews, included a considerable operational burden on both applicants and examiners, and a time lag between application and certification/skill visualization due to the annual examination schedule.
This system was developed to solve these challenges, maintaining evaluation accuracy while visualizing employees' specialized skills more timely and utilizing them for flexible job assignments to the right people. It structurally eliminates evaluation bias caused by hallucinations and textual expressions associated with generative AI, ensuring fair assessment of skills themselves.
System Overview
This system divides the evaluation process into two stages: a 'Quality Validation Phase' and a 'Skill Evaluation Phase,' and is equipped with a logical interlock that controls data flow to the subsequent stage based on the results of the preceding stage.
In the Quality Validation Phase, criteria can be variably set according to the diagnostic purpose. If criteria are not met, an information supplementation loop (additional questions) is executed up to the maximum number of attempts. If criteria are still not met, the process deterministically branches to interruption or forced transition.
System Features
① Multi-layered defense structure to eliminate AI bias and hallucination
[First Line of Defense] 'Quality Validation Phase' to block insufficient input
The evaluation process is divided into a 'Quality Validation Phase' and a 'Skill Evaluation Phase' to prevent AI from over-inferring context and forcing evaluations based on insufficient information. If the descriptiveness or specificity does not meet the criteria, the process does not proceed to the subsequent Skill Evaluation Phase, and the quality of the input content is ensured through additional questions.
[Second Line of Defense] 'Multiple parallel inferences' to eliminate hallucination
Instead of relying on a single AI inference result, AI evaluations are executed multiple times (N times) in parallel, and the results are statistically aggregated (e.g., majority vote or median) to eliminate factually incorrect information generation (hallucination).
[Third Line of Defense] 'Information concealment' to prevent evaluation bias
In skill evaluation, measurement (AI) and overall judgment (system) are completely separated. AI is only allowed to measure individual skills, and overall evaluation logic such as passing criteria is concealed from the AI. This prevents AI-specific evaluation biases like the 'halo effect,' where positive aspects in one area disproportionately elevate evaluations in other areas.
② Flexible adaptation to diverse evaluation systems
By parameterizing gate pass conditions, the system can flexibly adapt to various evaluation systems, such as numerical score evaluations and rank evaluations.
③ 'Deep-dive questions' like an interviewer
If an applicant's input content lacks specificity or essential information for evaluation, the system identifies the missing parts and presents them with reasons. This ensures that necessary information for evaluation is appropriately elicited, leading to more precise and fair assessments.
Regarding SCSK Group's Materiality
Towards realizing its management philosophy 'Creating a Future Full of Dreams Together,' the SCSK Group aims for sustainable growth with society, promoting 'Sustaina
This system ensures the validity and consistency of evaluations and achieves fair and flexible skill assessment through multiple AI verification processes and evaluation logic based on accumulated skill definitions. By introducing this system, SCSK will significantly reduce the man-hours required for evaluation, immediately visualize employee skills acquired through work experience, and contribute to further promoting human capital management by utilizing these insights for timely job assignments that foster employee growth and maximize their abilities.
Background and Purpose
SCSK has implemented the 'Specialty Certification Program' as a mechanism to support the continuous growth and career development of each employee by visualizing their specialized abilities across 18 job categories, 41 fields, and 7 levels.
Annually, skill levels are comprehensively assessed through application documents detailing employees' owned skills and practical experience, and interviews with applicants conducted by experts in each field acting as examiners. Detailed level definition documents, which specify the necessary skills for each job category and field, serve as long-term career development indicators for employees, and feedback from examination results is utilized by employees as a guide for self-improvement.
On the other hand, challenges with this examination process, involving application documents and expert interviews, included a considerable operational burden on both applicants and examiners, and a time lag between application and certification/skill visualization due to the annual examination schedule.
This system was developed to solve these challenges, maintaining evaluation accuracy while visualizing employees' specialized skills more timely and utilizing them for flexible job assignments to the right people. It structurally eliminates evaluation bias caused by hallucinations and textual expressions associated with generative AI, ensuring fair assessment of skills themselves.
System Overview
This system divides the evaluation process into two stages: a 'Quality Validation Phase' and a 'Skill Evaluation Phase,' and is equipped with a logical interlock that controls data flow to the subsequent stage based on the results of the preceding stage.
In the Quality Validation Phase, criteria can be variably set according to the diagnostic purpose. If criteria are not met, an information supplementation loop (additional questions) is executed up to the maximum number of attempts. If criteria are still not met, the process deterministically branches to interruption or forced transition.
System Features
① Multi-layered defense structure to eliminate AI bias and hallucination
[First Line of Defense] 'Quality Validation Phase' to block insufficient input
The evaluation process is divided into a 'Quality Validation Phase' and a 'Skill Evaluation Phase' to prevent AI from over-inferring context and forcing evaluations based on insufficient information. If the descriptiveness or specificity does not meet the criteria, the process does not proceed to the subsequent Skill Evaluation Phase, and the quality of the input content is ensured through additional questions.
[Second Line of Defense] 'Multiple parallel inferences' to eliminate hallucination
Instead of relying on a single AI inference result, AI evaluations are executed multiple times (N times) in parallel, and the results are statistically aggregated (e.g., majority vote or median) to eliminate factually incorrect information generation (hallucination).
[Third Line of Defense] 'Information concealment' to prevent evaluation bias
In skill evaluation, measurement (AI) and overall judgment (system) are completely separated. AI is only allowed to measure individual skills, and overall evaluation logic such as passing criteria is concealed from the AI. This prevents AI-specific evaluation biases like the 'halo effect,' where positive aspects in one area disproportionately elevate evaluations in other areas.
② Flexible adaptation to diverse evaluation systems
By parameterizing gate pass conditions, the system can flexibly adapt to various evaluation systems, such as numerical score evaluations and rank evaluations.
③ 'Deep-dive questions' like an interviewer
If an applicant's input content lacks specificity or essential information for evaluation, the system identifies the missing parts and presents them with reasons. This ensures that necessary information for evaluation is appropriately elicited, leading to more precise and fair assessments.
Regarding SCSK Group's Materiality
Towards realizing its management philosophy 'Creating a Future Full of Dreams Together,' the SCSK Group aims for sustainable growth with society, promoting 'Sustaina