Commerce21 Co., Ltd. conducted a generative AI training program called "Corporate Reskilling" for 67 participants from April to September 2026. The tool used was ChatGPT. The program consisted of nine sessions from April 9th (Thursday) to August 7th (Friday). In the final review session, 58 out of 60 participants (96.7%) reported using AI three or more days a week. At the final reporting meeting, examples were shared of how the process of reconciling data and updating management sheets, which used to take 2-3 days each month, could now be completed in just a few tens of minutes.

When the training began in April, the company was in the process of creating an environment where the AI could be used in daily operations. Within a few months, the phrase "I asked AI" started to come up naturally in regular departmental meetings. This article will introduce how an IT company that handles e-commerce systems expanded its use.

Company Introduction

Commerce21 Co., Ltd. (abbreviated as Commerce21)December 1999Established in [year], this IT company is headquartered in Nishi-Shinbashi, Minato-ku, Tokyo. Since its founding, it has independently developed systems for large-scale e-commerce and has supported the e-commerce business of major domestic companies as an SIer (Systems Integrator) that builds and operates e-commerce systems.

In recent years, we have expanded beyond our SI (Systems Integration) business to include consulting services. Working together with our client companies, we are committed to creating value for consumers and addressing value chains that prioritize value creation, thereby striving to create value for society.

One of the company's strengths is deepening relationships with customers by starting with data."Royal Customer DX Solution"On their official website, they proclaim, "Value can be created depending on the data."

The vision is"Fill the world with value created from information."On their official website, they explain the meaning as "creating data from information and creating valuable experiences from databases." One of the values they offer is "not letting it end as just 'work'." This training was attended by employees from a wide range of departments, including product development, management/human resources/information systems, planning and sales, and quality control.

Issues before introducing generative AI

At the start of the training, each participant conducted a task inventory, listing their own work. When the 66 participants' tasks across 14 categories were tallied, the most time-consuming was document and paperwork creation, accounting for approximately 201 TP3T of the total. This was followed by studying and information gathering (approximately 141 TP3T), system and design development and production (approximately 121 TP3T), internal regular meetings (approximately 101 TP3T), and email and message correspondence (approximately 81 TP3T). These top three categories alone accounted for approximately 461 TP3T of the total.

Tasks such as creating, researching, and compiling information accounted for a large portion of their work time. One participant spent two to three days manually reconciling monthly figures and updating management sheets.

The environment for using AI itself was also being set up in parallel with the training. I've heard from the person in charge that some people were able to start using it with confidence once the company officially distributed ChatGPT business accounts. In the survey conducted at the start of the training, the average number of tasks using AI among the 59 respondents was 2.58. The ways in which it was used also varied from person to person, ranging from extended search functions to creating email content.

Implementation Details - A unique initiative by Commerce21 Co., Ltd.

• A 9-part program
  The program ran from the startup meeting on Thursday, April 9, 2026, through four workshops, two implementation support meetings with interim reports, and a follow-up meeting on Friday, August 7. 67 participants took part, working in groups to review their work processes. A final report meeting for each participant was held on Friday, August 28, and video learning and chat-based question support continued until the end of September.
• We will implement corresponding tasks to address the heavy workload identified during the business process inventory.
  The workshop covered topics such as creating meeting minutes and emails, using GPTs to automate routine tasks with chatbots, creating tables in Excel, preparing documents and conducting research using Canvas and Deep Research. The workshop addresses all aspects of document creation, information gathering, meetings, and email correspondence, which were time-consuming during the inventory process.
• Support for utilization using regular tasks and "vague consultations" as examples.
  In the first support meeting, we covered the scheduling function that prompts weekly and monthly checks and reviews using ChatGPT, and a workshop on organizing ambiguous consultations into clarification questions, improvement suggestions, and decision-making criteria. Samples were prepared for each job type, such as technical and administrative roles, so that participants could try them out by adapting them to their own work.
• Codex work tailored to organizations with many engineers
  Since many of our employees are involved in system development, the second support meeting included not only the latest information and Q&A, but also a mini-workshop on the differences between ChatGPT and Codex, and how to perform data aggregation, document compilation, and code modification using Codex. In the follow-up meeting, we demonstrated the process of creating sales proposal materials using Codex.
• Develop themes based on feedback from participants.
  We incorporated the "themes you'd like us to cover in the future," gathered from the survey, into the content of the second support meeting. Participants have also commented that the latest news and introductions to new features shared at each meeting provided them with an early opportunity to experience the features.

Benefits of using generative AI

Among the examples shared at the final reporting meeting, the most significant change was in one participant's monthly data reconciliation and management sheet updates. This process, which previously took two to three days of manual work, was completely replaced by using AI to generate the sheets. The same task can now be completed in just a few tens of minutes. The participant says that the time saved has allowed them to tackle tasks they previously couldn't manage.

Case study of one participant: Monthly reconciliation and updating of management sheets(Values shared by the person in charge at the final reporting meeting)

BEFORE	2-3 days every month
AFTER	several tens of minutes
Those who use AI three or more days a week (58 out of 60 respondents)	96.7%
Self-assessment level 3 or higher (59 out of 60 respondents)	98.3%
Number of tasks utilizing AI (average of 60 respondents)	6.32 items
Weekly reduction in working hours (self-reported, with a maximum of 10 hours)	7.27 hours(Average of 60 respondents)

Its use has become a part of daily life. In the review on August 7th (Friday), out of 60 respondents...58 people (96.7%) use AI three or more days a week, and 45 of them use it five days a week.They answered that they use it. In the self-assessment, 59 people (98.3%) selected Level 3 "Usable to a limited extent" or higher, and of those, 47 people (78.3%) selected Level 4 "Used to improve work efficiency."

The average number of tasks utilizing AI was 2.58 in the initial survey (59 respondents), while the average number in the final review (60 respondents) was 6.32.

* The figures are self-reported values from the participants. The number of tasks utilized is a comparison of the average responses from the start (April 9, 2026 (Thu) survey, 59 respondents) and the final point (August 7, 2026 (Fri) review survey, 60 respondents), and the respondents were not the same at each point. The level of self-assessment was selected by the participants themselves from the given options. The weekly time reduction is a value calculated by counting responses exceeding 10 hours as 10 hours.

AI utilization advanced in the higher-level business processes of business inventory.

Tasks that took up a lot of time during inventory in April	Percentage of the whole	Those who listed tasks they have entrusted to AI
Document and material creation	Approximately 20%	34 out of 60 people
Studying and Information Gathering	Approximately 14%	22 out of 60 people
Internal company meetings (regular meetings)	Approximately 10%	18 out of 60 people
Email and message support	Approximately 8%	19 out of 60 people

* The right column shows the number of people who answered the question "Of the tasks you inventoryed in April, which ones are now being handled by AI (multiple answers allowed)" in the reflection survey, categorized by keyword and assigned to a specific task area. If one person listed multiple tasks, they were counted separately. This does not mean that the time spent on each task decreased by the same percentage. Internal meetings are categorized as including participation and management, and the main tasks mentioned were the creation of meeting minutes. The task inventory is based on data from 66 people as of April 2026.

In inventory management, the use of AI for the most time-consuming tasks has become widespread. In particular, document and paperwork creation was cited by more than half of the respondents. Even in specialized tasks, there were comments that the speed of work has increased as AI now handles everything from research to document creation, and that it is being used for code reviews and design document creation.

The following is a comparison of how work is carried out before and after the implementation.

Before implementation | Time-consuming for verification and creation.

• The monthly reconciliation of figures and updating of management sheets are done manually.2-3 days)
• Research materials are created after personally verifying numerous documents and sources.
• Business knowledge resides in each individual's hands, making it time-consuming to share.
• If you don't know something, first try searching for it yourself.

After implementation | Time is created that is not spent on "tasks".

• By changing the process to have AI create the table, the same task can be performed.several tens of minutes(An example from one participant)
• AI handles everything from research to document creation.Speed increased and fewer omissions and errors occurred.Some people feel this way.
• Share your knowledge with GPTs.The ability to expand horizontally has increased.Some people say that
• First, consult the AI, then verify the source and final confirmation yourself.

These changes are also reflected in the free-response sections of the reflection sheets. First, there are comments that the way people interact with AI has changed.

"Experiencing firsthand the benefits of using generational AI in my own work changed my perception of AI itself from something convenient to a tool that should be utilized. Before I knew it, I was using it on a daily basis, which surprised even myself."

From the training reflection sheet

When using it becomes part of daily life, individual ingenuity begins to transform into an organizational asset. Here are some comments regarding the sharing of knowledge using GPTs:

"Until now, we relied on each individual's knowledge and were unable to effectively output it, but by using GPTs, which are easy to create, we have been able to share and disseminate knowledge much more quickly."

From the training reflection sheet

And the more you use AI, the more important it becomes to know how to maintain a healthy relationship with it.

"Furthermore, there's a growing awareness that humans should be responsible for verifying the source and making final confirmations, leading to a greater reliance on AI rather than blindly trusting it, and instead viewing it as just one option among many."

From the training reflection sheet

At the final reporting meeting, participants shared that they had learned through training how to prompt users to indicate the source of information, and that those receiving inquiries were now able to say, "Please think about it yourself first, then consult ChatGPT and bring it to me." The question is no longer whether or not to use AI, but how to use it while verifying its effectiveness, and this is starting to become a common topic of discussion within the company.

Future outlook

Now that most of those who responded to the review are using AI on a daily basis, the next point of discussion is how to connect individual use to departmental and company-wide systems. The final report shares the following points:

• Sharing how to use information to ensure its accuracy
  In the July survey, the most frequently cited obstacle to adoption was "concerns about the accuracy and reliability of the generated information." Since there are still individual differences in how fact-checking is conducted, it would be advisable to standardize the verification process within the company.
• Establishment of scope of use and operational rules
  How should we define the scope of tasks that will use AI, and how should we manage its operation when combining it with other tools? Now that its use has expanded, one of the next points to consider is creating a system that supports it with rules.
• Tool integration, automation, and deployment to development.
  The areas participants most frequently mentioned as areas they want to deepen their knowledge of next were tool integration and automation, and development and engineering using tools like Codex. Now that general-purpose usage has become established, interest is shifting to more practical and technically-oriented applications.

Through a six-month initiative, Commerce21 Co., Ltd. progressed from the "testing" stage to the "using in daily operations" stage of generating AI. One of the results achieved during this process is that the matching process, which used to take 2-3 days each month, has been reduced to just a few tens of minutes, and 96.7% respondents in the review survey now use AI 3 or more days a week.

"Filling the world with value created from information"—for this company, which works daily to create value from data, AI is becoming a new tool to reduce "tasks" and reclaim time for thinking.