PhonePe’s data science team acts as one of the foundational pillars of its payments service to millions of people across India. Founded in 2015, the company became the most widely used payments service, amassing 40% of the total UPI payments with 925 million transactions overall in October 2020. Being trusted by 250 million users, PhonePe extensively leverages machine learning to ensure customer trust in a highly regulated industry. A strong Data Science culture is a key pillar for the rising popularity and leadership of PhonePe in this highly competitive space.
To peek into the data science culture of the company, Analytics Drift got in touch with Kedar Swadi, head of data sciences at PhonePe. Kedar shared interesting insights into the data science team of the Walmart-owned payments company as well as shared his opinions on best practices for aspirants to flourish in their careers.
How Is Machine Learning Leveraged At PhonePe?
Machine learning not only helps PhonePe users and partner merchants enhance customer satisfaction but also is leveraged within the organization to optimize business processes for increasing revenue and reducing costs. “Data science is at the core of a lot of decisions that we make at PhonePe; we utilize patterns in users’ payment history to remind them about the right payment at the right time,” says Kedar. It may be a payment to another user or a bill that one should not miss or recharge to keep the continuity of services. Besides, with superior machine learning techniques, PhonePe ensures that it identifies potential fraudulent users as early as possible to maintain a healthy payments ecosystem.
PhonePe’s Data Science Culture
“One of the many great things about data-driven PhonePe is its data science culture. We thrive on taking up hard problems while limiting the risks and experimenting to arrive at better solutions,” explains Kedar. The decision-makers back the data science efforts to enable the team to blaze a trail in various payment aspects. Data scientists at PhonePe are provided with the necessary time to explain the problems, gather information, and extract value from multiple data. With this, PhonePe allows innovation and accepts failures as a part of data science initiatives to keep the team motivated during setbacks.
While following the right set of principles to drive the team is one aspect to thrive in the competitive market, having the right professionals who fit in the culture is equally important. And since it takes years to understand the basics, tools, and techniques for a data science job, PhonePe relies on professionals who have two to three years of experience in the field. However, Kedar believes there are always exceptions, even for someone who wants to change careers but may have to start at the entry-levels and build the required skills.
And when asked about the critical skills he seeks in data scientists while hiring, Kedar said that a strong background in algorithms, basic statistics, optimization, and programming skills is a prerequisite. Although some experience in the application of multiple machine learning algorithms helps, he believes new algorithms can always be learned on the job if one has obtained foundational skills. And since data science is also a communication-heavy discipline, Kedar stressed the importance of effectively communicating, whether it is to understand the problem or explain the results. Apart from the aforementioned skills, he evaluates applicants for the cultural fit-aspirants who can work in hierarchy-less organizations like PhonePe.
But, how does PhonePe hire the desired data science professionals? Hiring data scientists is a massive challenge for organizations due to the talent gap in the market. As a workaround, PhonePe looks out for good people and then provides them with the support and opportunities to grow and get better. “A lot of emphasis is laid on teamwork and supplementing each other’s skills. We look for people who show other valuable traits like curiosity, perseverance, self-motivation, and desire to excel,” says Kedar.
Do Data Scientists With Ph.D. Perform Well Over The Self-Learned?
According to Kedar, a Ph.D. implies a solid five or more years spent understanding a topic and theory behind the techniques. This helps learners assimilate the strengths and weaknesses on a wide range of machine learning methodologies, enhance the ability to formulate and solve problems independently, and train to deal with failures. On the other hand, self-learning with MOOCs focuses on specific skills in a reasonably short time and a fast-paced environment. Therefore, a Ph.D. has an advantage but a few years of on the job experience by a self-learner usually reduces the gap.
Advice To Aspiring Data Scientists
Data science is about handling data, drawing insights from it, and communicating in a meaningful and actionable way with the various functions in an organization. To carry out all of these, one needs to have technical competence, domain expertise, and most importantly, the right attitude.
“When it comes to technical competency, you need to have a strong foundation in algorithms, mathematics, and statistics. Learning new techniques when needed will become easy when you have acquired solid fundamentals. You should also be good at programming, know how to handle large amounts of data efficiently, and create scalable pipelines to work with ever-increasing amounts of data,” mentions Kedar.
“But domain expertise is acquired through experience. For instance, if you do not have sufficient knowledge about radiology, you would struggle to understand the data, problems, and other nuances of the field. Mastering even the sub-domains like banking, finance, insurance, investments and trading under the BFSI sector will take years. Consequently, steadily improving the data science skills in conjunction with the domain expertise will be the key to a successful career.”
“Finally, as a data scientist, one needs to have the right attitude to believe in the process and not just implement glamorous machine learning techniques. 80% of data science is manual work of looking, cleaning, processing data. This requires humility to work on non-glamorous aspects of data science. The tenacity to sift through a massive amount of data, the ability to deal with failures, and the strength to learn and get better are equally essential for data scientists,” concludes Kedar.