Restarting from scattered PoCs and a 12% usage rate.
Achieving a 68% company-wide active usage rate.
A governance-driven internal AI implementation that achieved a 68% company-wide active usage rate
Industry
Precision equipment development and manufacturing
Implementation scope
Company-wide
(Development, production, sales, legal, general affairs, and quality assurance)
Implementation period
Diagnosis / PoC: 2 months → full implementation → company-wide rollout completed in 7 months
BACKGROUND
Rebuilding the AI strategy had become urgent.
Company B introduced a general-purpose LLM-based internal AI three years earlier, but adoption failed because employees felt they could not write prompts or were unsure whether confidential information could be entered. Usage dropped from 70% three months after launch to 12%. In addition, each division began ordering chatbot PoCs separately, causing AI-related licenses to proliferate. The director responsible for DX repeatedly pointed out that the team could not explain the return on investment numerically, making it urgent to rebuild the AI strategy. CLAVI Mining was selected as the platform to consolidate this “ungoverned proliferation of AI.”
Challenges before implementation
The issue facing Company B’s management was not simply “AI that was not used,” but “AI proliferating without governance.”
Poor adoption caused by dependence on IT literacy
The existing general-purpose AI only produced results for employees who could write good prompts. As a result, it remained a tool used only by some younger engineers. Adoption did not progress in back-office departments such as legal, general affairs, and sales administration, leaving the company far from achieving company-wide DX.
Lack of governance around information leakage risk
Several audit findings revealed that employees had unintentionally pasted design drawings, unreleased product specifications, and pricing information with clients into general-purpose AI tools. At management meetings, many warned that promoting AI use under these conditions posed a serious risk, and restrictions on AI use were even considered.
Rising costs and loss of control from scattered PoCs
Five business divisions had each contracted separate chatbot PoCs with different vendors. Annual license costs totaled about ¥32 million, yet none had reached production. The information systems department could not fully track each division’s individual contracts, and ID management and access revocation for departing employees were not assured.
Difficulty quantifying AI as a management KPI
At every quarterly board meeting, the team was asked, “What is the impact of our AI investment?” However, there were no unified metrics for usage or work-hour reduction, making it difficult for the DX promotion department to explain results.
At management meetings, many warned that promoting AI use under these conditions posed a serious risk, and restrictions on AI use were even considered.
Reasons for selection
The reason Company B’s director of DX ultimately selected CLAVI Mining was not merely its performance as an AI tool, but its fit as a company-wide governance platform.
First, reliability as an “AI that does not make mistakes,” based on patented technology.
If generative AI gives incorrect answers in legal or contract-related work, it can directly damage corporate credibility. The 78% reduction in misinformation output through multi-layer feedback control and a fact-checking policy engine became decisive evidence for fulfilling accountability to the board.
Second, company-wide rollout capability regardless of IT literacy.
The UI and customer-success support enabled field managers, legal staff, and sales administrators to produce results from day one without prompt-engineering knowledge. This matched the management policy of retrying AI as a company-wide DX initiative.
Third, audit and internal-control readiness through transparency logs and verification APIs.
The design records evidence for every response and can be incorporated into SOX and J-SOX audit scopes, which was an essential requirement for a listed company.
Fourth, minimizing information leakage risk through a dynamic prompt sanitizer.
The mechanism detects and blocks inputs containing confidential information in real time, structurally preventing a recurrence of incidents where design information was leaked to external AI due to employee carelessness.
Fifth, automatic aggregation of management KPIs through monthly reports.
Visualizing usage rates, inquiry reduction rates, and reduced work hours solved the long-standing accountability challenge around AI investment for the director in charge of DX.
Results after implementation (4 months after launch)
Four months after implementation, the following numerical improvements were confirmed.
Company-wide active usage rate
The monthly active usage rate reached 68%, a 5.7x increase from the previous system’s 12%. In particular, usage in back-office departments such as legal, general affairs, and HR increased twelvefold compared with the previous system, significantly changing the perceived value of company-wide DX.
Contract and specification review workload
Review work for contracts and specifications in legal and general affairs was reduced from an average of 3.5 hours to 1.2 hours, a 66% reduction. This created the equivalent of approximately 1.1 person-months of work capacity annually.
AI license cost consolidation
By consolidating the scattered PoC licenses across divisions into CLAVI Mining, annual AI-related license costs were reduced by about ¥24 million. After consolidation, the information systems department could centrally manage IDs and permissions.
Audit response workload
During annual internal audits, extracting and analyzing company-wide AI usage logs previously required two person-months. With CLAVI’s transparency logs and reporting functions, the process was completed in a few hours, greatly improving the internal audit department’s evaluation.
Employee NPS
The internal NPS score for AI tools improved significantly from -12 before implementation to +47 afterward. Comments such as “AI that understands us” and “AI we can safely use for work” made up the majority of qualitative feedback.
Comment from the Director of DX
AI should be evaluated not by “intelligence,” but by “governance.”
“What we learned from the failure of the previous system is that AI should be evaluated not by how smart it is, but by whether everyone can use it safely under proper governance.
CLAVI Mining combines reliability through patented technology with governance functions essential for listed companies, which allowed the investment proposal to pass smoothly at the board meeting for the first time.
Next, we are considering rollout to three group companies and multilingual deployment for overseas sites.”
Director of DX
Director of DX