Case Study – Analytics and Machine Learning

1) Customer Recency, Frequency and Monetary Analysis

Goals & Approach

  • To gain insights into customer behavior throughout their life cycle — from acquisition, Onboarding, servicing, cross-sell and portfolio management, to retention and delinquency management using data science technologies.
  • Key Insights & Highlights

  • Recency, a frequency and monetary (RFM) analysis for retail customers for truly personalizing customer experience through Nano targeted merchant offers.
  • It resulted in average spend per customer to increase by 29%, while spend for the targeted customer base increased by 44%, which contributed 27% growth to overall portfolio spend.
  • Clustering techniques (K-Means) were used to segment the customers for promotion



  • 3) Commercials: AML Rules

    Goals & Approach

  • To provide analytics solution for flagging financial transactions that may violate legal or compliance frameworks and building some Anti Money Laundering rules.
  • Key Insights & Highlights

  • All the current account data gathered from bank CASA ledger database.
  • A GBM Classifier model created on the past data behaviour for all the transactions going through these current accounts.
  • The system was able to capture more 95% transactions which were identified as ML transactions on the test data.
  • The Amlgo data science team were supported by Amlgo domain experts with a detailed understanding of Banking processes.


  • 5) Risk & Litigation Model

    Goals & Approach

  • To provide analytics solution for avoiding a claim going into the litigation based upon the features of a claim or imputs provided by claim handlers, determining the litigation propensity for general damages claims.


  • 2) Customer Data Management Goals & Approach

    Goals & Approach

  • Banks are obliged to collect, analyze, and store massive amounts of data. But rather than viewing this as just a compliance exercise, using machine learning and data science tools can transform this into a possibility to learn more about the customers to drive new revenue opportunities.
  • Key Insights & Highlights

  • Identification of specific customers for newly introduced credit card and the conversion rate was more than 30%.
  • Being armed with information about customer behaviors, interactions, and preferences, the average spend on the card was more than USD 1000 per month.
  • Further analyzed the spend data and collaborated with around 10 vendors for commission-based revenue growth.

  • 4) Insurance: Anti-Fraud Model

    Goals & Approach

  • To improve fraud capture rate and conversion rate for third party bodily injury claims, Utilised machine learning techniques to predict the chances of fraud claim case.
  • Key Insights & Highlights

  • Implementation: Decision Tree Algorithm used to develop low speed impact and stage incident fraud models.
  • After that, developed a Power BI report for quick reference for claim handlers.
  • Model Outputs: Fraud propensity scores categorized into risky segments such as high, medium and low.






  • Key Insights & Highlights

  • Implementation: Random Forest Algorithm used to develop the propensity scores – low, medium and high.
  • After that, developed a Power BI report for quick reference for claim handlers.
  • Model Outputs: Litigation propensity scores embedded in a application with configurable parameter by users.
  • Case Study – Regulatory Reporting

    1) CECL Accounting and Disclosures Implementation

    Goals & Approach

  • The US variant of IFRS 9 impairment is called CECL (Current Expected Loss model).
    The overall target is to adapt the IFRS9 framework to cater as CECL in respect to configuration, calculation and disclosure reporting possibilities for a front-end and back-end perspective.




  • Key Insights & Highlights

  • Implemented CECL configuration, risk models, scenarios and disclosure reporting that includes CECL calculation specifics.
  • The solution is having management reporting capabilities to enable sensitivity analysis and top-down (post-calc) stress testing, both through Business Analytics.
  • Several underlying features and a bit more work from an administration point of view completed to build the end to end solution.