DataRobot Case Studies Zidisha is Transforming Lives with DataRobot
Edit This Case Study Record
DataRobot Logo

Zidisha is Transforming Lives with DataRobot

DataRobot
Analytics & Modeling - Predictive Analytics
Finance & Insurance
Education
Procurement
Predictive Quality Analytics
Fraud Detection
Data Science Services
Zidisha, a non-profit online microlending community, aims to transform the lives of people in some of the poorest countries by offering microloans to create businesses, attend school, or improve their living conditions. However, every loan carries the risk of default. Traditional lenders have found ways to identify, quantify, and price default risk, with higher risk loans attracting higher interest rates. The work of assessing risk commonly falls to a loan officer and the costs are passed on to the borrower. In developed economies where loans of thousands or hundreds of thousands of dollars are common, these costs can be comfortably absorbed without undermining the case for taking a loan, but this is not the case in developing countries. Employing a loan officer to assess default risk for a microloan results in interest rates as high as 40%, undermining promotion of economic development. Zidisha's challenge was to improve levels of repayments by identifying applicants most likely to be high-risk borrowers.
Read More
Zidisha is a non-profit online microlending community that connects borrowers directly with lenders. It promotes crowdfunding of micro-loans for people in some of the poorest countries in the world to start and grow a business or fund their education. Zidisha does not charge interest on loans; instead, borrowers pay a 5% service fee to cover the cost of transferring and administering the loan. With Zidisha, entrepreneurs get access to business loans on flexible terms and at an affordable cost, allowing them to keep a majority of their profits and invest them back into their businesses or use them to support their families. Zidisha circumvents traditional approaches to identifying, quantifying, and pricing default risk, removing the need for loan officers or bank specialists. In place of due diligence undertaken by a loan officer, Zidisha fosters direct relationships between borrowers and lenders.
Read More
Zidisha partnered with DataRobot to develop and deploy machine learning models that radically improve the loan application and screening processes. A customer facing data scientist (CFDS) at DataRobot suggested Zidisha would benefit from two predictive models: one to detect fraudulent applications and a second to identify applicants with a high propensity to default on their loan. Protecting their lenders’ money is fundamental to Zidisha’s long term success, as it makes more loans available to trustworthy borrowers who truly need them and increases the rate at which money is recycled to other worthy borrowers. Julia and her colleague worked with the DataRobot team to integrate the platform with their systems, and simply read the DataRobot user docs to start developing their own predictive models for Zidisha. The two models created by Julia and her colleague in less than two weeks have profoundly improved the percentage of loans repaid to lenders on Zidisha by reducing loan defaults by 5%.
Read More
Zidisha's models are directing more capital to people who otherwise wouldn’t be able to afford to attend university, send their children to high school, or start a business that will improve their family’s standard of living and boost the local economy.
By reducing the risks of losing lenders disheartened by poor rates of repayment and keeping capital in the hands of reliable borrowers, DataRobot is helping Zidisha to achieve its mission and improve the lives of thousands of people across the world.
Zidisha admits greater numbers of borrowers but only advances small loans to individuals assigned a high risk score.
Zidisha advanced loans worth $500,000 in a recent month.
If the DataRobot Enterprise AI platform helped Zidisha lift loan repayments by only 1%, the platform would pay for itself.
Zidisha's machine learning model had a 5% lower default rate measured in dollars repaid.
Download PDF Version
test test