Internships usually give students a chance to explore classroom knowledge in the working world. And for The NUS MBA candidate Keshav Jain, it was the perfect opportunity to do a little more — applying machine learning to effective sourcing of prospective founders during his internship with early-stage venture capital firm Antler.
Keshav, who hails from the financial sector, saw how machine learning improved risks management and trading processes from work. Even with limited coding experience, he built a system that predicts the success rate of prospective start-ups’ founders. This is expected to help Antler in its sourcing strategy. He shares with BIZBeat his experiences with machine learning at Antler.
Machine learning in the industry
It has always been fascinating to see the insights that a well-written code can generate from
raw data. In the field of Finance, Artificial Intelligence (AI) and Machine Learning (ML) have been growing in prominence over the last few years, with applications in several niche fields, including credit card default prediction and the creation of risk-return optimised portfolios.
Even trading desks have now started to rely on algorithms utilising AI concepts to create and execute complex trading strategies in real-time. Prediction is one of the critical aspects taken into consideration in multiple areas of banking and investments. Prediction of returns, defaults or values of various parameters, including macroeconomic variables, is crucial in decision-making. Prediction may lead to binary categorisation in the following examples. A borrower may be predicted to default or not to default. Supervised ML algorithms rely on several sophisticated techniques, such as Logistics Regression and K-Mean Segmentation, to predict and classify data based on an initial training set of data.
The NUS MBA candidate Keshav Jain
Starting a new journey in venture capital
After attending the Venture Capital module taught by Professor Ng Weiyi in my first semester, I developed a keen interest in the field. So, to explore it further, I started to look out for part-time internships in the Venture Capital (VC) space and came across an internship opportunity at Antler. Antler is a global early-stage venture capital firm that brings exceptional individuals together to build tech companies from the ground up. After speaking with a few people at the company and finding the opportunity exciting, I applied for an internship position and was selected after screening tests and interviews.
When I started my journey with Antler in May, I quickly identified how ML algorithms could be leveraged to predict founders’ success. In addition, the insights generated could also help narrow down the most important attributes in determining the chances of success of a founder. Examples of these individual attributes include education, work experience, past entrepreneurship experience, domain expertise, and skillsets. The goal of the ML algorithm was to identify a relationship/function between these attributes and the chances of success, which helps the prediction process.
Transforming venture capital with software
While framing the problem statement was easier, writing a code was challenging because I had not done any programming since high school. So, after some research, I decided to use Python because it’s user-friendly, easy to learn, and has a lot of libraries for statistical modelling and machine learning.
I utilised the past data of founders as the training set and developed a regression-based ML model to link the founder attributes with chances of start-up success. The model showed a healthy accuracy of test data. However, the actual accuracy of the model is expected to be much lower, given the limited training dataset used in building the model.
Yet, the entire modelling exercise using ML provided two key outputs. The first output is to help identify some of the important and “not-so-important” personal attributes in evaluating the success of a few founders over others. While some of these attributes are more obvious than others, there were a few unexpected results. The other key milestone achieved through this initial ML model is that it can predict the success probability. The precision of the prediction is yet to be evaluated for prospective founders, and accuracy is expected to be low. However, the model does provide a prototype that can be trained and re-engineered based on larger datasets, leading to a more accurate and robust version.
Learning experience
As an MBA internship experience, the entire exercise of working on this project was both exciting and very enriching. It was all possible due to the support and encouragement from the wonderful team at Antler. I found the team environment very friendly and collaborative right from the beginning. I also got the opportunity to interact with senior management at the firm, who shared constructive insights. Even though I was working remotely due to the pandemic restrictions, I always connected with the team.
Though I was faced with the challenge of balancing my internship work with my academic load of core courses, a consulting project and an additional elective during that semester, the flexible working arrangement and open culture in my team helped a lot. In addition, for someone like me who had spent the last seven years of my pre-MBA journey in banking and finance, the ability to experience the VC industry from the perspective of advanced technology and AI was mesmerising. While I learnt a lot about the industry, my key learning was how advanced data analysis techniques like Machine Learning could find their applications in industries as novice as Venture Capital. And for someone who hasn’t coded for a decade, well, it was even the more fascinating!
