Data Science Program for a Major Financial Group
Machine learning models and a BI system deployed across securities, foreign exchange, lending, and insurance.
Challenge
A major Japanese financial group spans several business segments: securities, foreign exchange, lending, and insurance. Data was spread across business lines, risk review and marketing analysis relied heavily on manual work, and each department defined its numbers differently. The group did not need a single model. It needed a data science and BI system that could go live in several business segments at once, and that business teams could use on their own.
Solution
Scope of work: Data science modeling across the group's business segments, group-wide BI rollout, digital marketing analysis, and training of in-house data talent.
- Risk control and fraud detection: A fraud detection model built with machine learning (an automated modeling platform) automated the review process, saving more than 20 hours of manual review per month. A credit risk ranking model supports risk assessment for lending.
- Customer and market modeling: Identifying high-LTV securities clients, researching models for short-term FX movement forecasting and trade automation, and analyzing call connection rates and TV ad effectiveness for the insurance business, with model outputs handed directly to business teams.
- Group-wide BI rollout: Tableau / Power BI rolled out across the group, with automated daily reports (saving another 20+ hours per month), so every business line reads data by the same definitions.
- Digital marketing analysis: Website UI/UX and marketing improvement proposals based on Google Analytics and A/B testing.
- Cloud and AI in practice: POCs for analytics and machine learning projects on AWS, Azure, and GCP, plus in-house AI / big data training to build a data-driven culture.
- Technical due diligence: Technical due diligence support in English for overseas tech venture investments.
Results
- Live in 4 business segments: Data analytics and machine learning deployed in securities, foreign exchange, lending, and insurance.
- Staff time freed up: Fraud detection alone saves more than 20 hours of manual review per month, and daily report automation saves another 20+ hours.
- A system, not a one-off: From models and BI tools to training and cloud environments, the group now has data capabilities that keep running on their own, not a one-time model that stops once delivered.
See other case studies
Eight platform-wide sales events in over two years with zero major incidents, load tested at up to 8 times normal traffic.
Video data and transaction data connected across about 500 stores, cutting the investigation of a suspicious transaction from a week to a day.
