border border

Optimising Credit Scoring and Risk Assessments using Alternative Operational Data

Team members

James Gan Sheng Wei (ESD), Reuben Tan Kunhan (ESD), Karan Nair (ESD), Sean Michael Lim Zihan (ISTD), Koh Yi Zhi Elliot - Ezekiel (ISTD), Pham Trung Viet (ISTD), Sidharth Praveenkumar (ISTD)


Georgios Piliouras, Natarajan Karthik Balakrishnan, Matthieu De Mari

Writing Instructors:

Rashmi Kumar

Teaching Assistant:

Siddharth Kumar

asset 2 3x 100

Problem Statement

Small and underbanked businesses are unable to get loans due to a lack of sufficient credit history, resulting in them being perceived as high risk. Furthermore, ShuttleOne's current attempt to look at operational data as a solution is impacted by the inefficiencies of manual data entry and manual loan approvals. This is due to ShuttleOnes inability to find an effective and efficient way to approve loans.



Current Process Visualisation

current process



Limitations of Current Process

ShuttleOne automated current processes have a success rate of 5%. Thus, 95% of the work is done manually. The low success rate is due to certain issues shown on the right.



asset 1 2x


Our Mission

Developing a credit scoring system utilizing Optical Character Recognition to obtain parameters from bank statement data and additional parameters from the company database to generate a credit score for the company to determine the suitability of a client receiving a loan from the company.


Our Solution

asset 2 2x




Integrating our solution into Shuttleone's platform

Our solution: Effective Optical Character Recognition (OCR) and Data-driven Credit Scoring Model programs packaged in the form of Application Programming Interfaces (APIs). With this, the solution is integrated into Shuttleone's platform to perform the automated loan approval as shown in this video.


New System Workflow



Software Used

asset 8 1 25x





As compared to ShuttleOne's previous model where only 7% of bank statements were being converted automatically, our model has a vast improvement that is able to convert 92% of bank statements. The credit scoring model created by the team also uses 10 features as compared to 1 feature being utilize previously

asset 9 1 25x

In collaboration with:




student James Gan Sheng Wei Engineering Systems and Design
student Reuben Tan Kunhan Engineering Systems and Design
student Karan Nair Engineering Systems and Design
student Sean Michael Lim Zihan Information Systems Technology and Design
student Koh Yi Zhi Elliot - Ezekiel Information Systems Technology and Design
student Pham Trung Viet Information Systems Technology and Design
student Sidharth Praveenkumar Information Systems Technology and Design
border border