Rolled out generative AI across a financial-services company under financial-grade controls, cutting operational workload and cost through automation
Background
The company wanted to use generative AI across the whole organization to raise productivity. As a financial-services business, however, it had to strictly observe Japan's APPI (personal-data law), internal regulations, and confidential-information handling — uncontrolled use would have been a compliance risk.
Constraints
- Alignment with Japan's APPI and internal regulations
- Managing the risk of confidential data leaving the company
- Conformance with J-SOX IT general controls (ITGC) and the FISC security guidelines
- Responding to internal audits and remediating findings
Approach
Chose to 'enable safe use' rather than prohibit. Established usage guidelines and designed permissions and logging first, then built a platform that monitors 100% of input prompts. With controls in place, rolled out business automation across the company in stages.
Implementation
Built and operate multiple generative-AI business systems around the Claude / OpenAI APIs and AWS: (1) fully automated specification generation for existing systems, (2) call-recording transcription with LLM-based first-pass evaluation, (3) full monitoring of employee prompts to generative AI, (4) first-response drafting for customer inquiries, (5) an answer-generation harness for J-SOX ITGC / FISC security-guideline assessments, and (6) automated research on unlisted companies.
Results
- Established company-wide generative-AI use without loosening controls
- Cut operational workload and cost by automating multiple workflows
- Made confidential-data handling risk visible and manageable through full prompt monitoring
- Addressed and remediated internal-audit findings
Learnings
Generative-AI governance is not about banning — it is about designing for safe use. Control and adoption are not a trade-off: putting logs, permissions, and monitoring in place first makes both possible.