Discover this year's submissions to the Dataiku Frontrunner Awards and give kudos to your favorite use cases and success stories!READ MORE

PT. iZeno Teknologi Indonesia - Creating a Data-Driven Solution for Customer Segmentation, Fraud Detection, and Credit Scoring

Name:

Deddy Johari, Director of Technology
Natalia Kemal, Senior Project Manager
David The, Analytics Consultant

Country: Indonesia

Organization: PT. iZeno Teknologi Indonesia

iZeno Pte Ltd was established in 2003 and provides technology and business consulting, software solutions, and system architecture design, integration, and management for Medium and Large Enterprises. Today, iZeno operates in five major cities in ASEAN and has provided services to more than 300 enterprise customers across the Asia Pacific. We have a highly experienced management team and best-in-class professionals who understand your business and technology imperatives. iZeno’s professionals leverage extensive industry and technology domain experience and flexible tools and methodologies to successfully deliver projects on time and on budget.

Awards Categories:

  • Partner Acceleration

 

Business Challenge:

The end user's main challenge is that the business data model differs from conventional data, where the customer is a small business group within different lower-income regions and groups. The team is also new to analytics modeling and was recently tasked with analytics implementation that can be used to support market frontliner in making decisions on approval and using a different strategy to grow the business. The end user’s current infrastructure is in a conventional RDBMS data warehouse already running for reporting purposes. This data is not being maximized to generate more revenue or used to better understand their customer and give better customer service.

 

Business Solution:

We architected the whole solution using real-time data streaming on Confluent into a semi-structured database MongoDB cluster as a single source of truth. This data is then analyzed using Dataiku, where all processing is pushed down into Spark running on the same MongoDB cluster hardware for faster analysis results.

We implemented three use cases in total: customer segmentation, fraud detection, and credit scoring. Due to the different abilities to combine rule-based and machine learning-based modeling, we chose Dataiku as it has the open capabilities to support multiple processing engines and read from different data sources. We also took a flexible and dynamic preparation model feature, combined with superior operational capabilities as a strong consideration.

Customer segmentation is implemented based on lender’s profile data where we need to split the data by different geographic regions due to various economic scales. We are able to segment the customer into seven different clusters and present this result for better decision support on growing the business.

Fraud detection is implemented by grouping lenders into smaller groups of people and using payment data, due date, and lending frequency to identify the potential fraud that will help the account service reduce the risk of lending default.

Credit scoring implementation uses a more complex FICO model learning that takes different parameters such as lenders’ payment history and their profile to pinpoint the scoring. Based on a specific defined range, the scoring generates five different categories to decide a more accurate decision process.

 

Business Area: Financial Services Specific

Use Case Stage: In Production

 

Value Generated:

Technically, the amount of time spent to run the data processing improves significantly by removing the previously manual step done for data preparation before analyzing the data manually. Users are able to automate this process and spend more time analyzing and building the model, instead of grouping and manipulating the data.

For each learning cycle, using spark processing maximizes the time by up to 30% for each learning cycle and enables future speed expansion by adding more processing nodes.

From a business perspective, customers' reach is improving with a visible uptrend in generating more lending services with fact-based segmentation. The end user is also able to have a 50% faster rejection time on fraudulent transactions.

 

Value Brought by Dataiku: 

As the first adoption into the Analytics and Machine Learning journey, Dataiku helps the customer accelerate their journey and ensure the platform can be operated efficiently at the same time. We have also ensured end users’ continuous adoption by investing less in programming and more in business use case expansion.

Dataiku implementation also provides transparency on each data modeling and learning process so that it can always be enhanced further and make the knowledge transfer easier. A large community of Dataiku users provides end user’s with resources for upskilling, exploration, and troubleshooting on the shared platform. The targeted learning path from Dataiku academy is also helpful for the end user’s to onboard faster into their analytics journey.

 

Value Type:

  • Reduce risk
  • Increase trust
Share:
Version history
Last update:
2 weeks ago
Updated by:
Contributors