PhonePe is one of the largest fintech players in the country with 304 million users spread across 12,000 towns and 20 million stores. The digital payment company’s  data science team is engaged in fine-tuning marketing campaigns, optimising operating costs, deep personalisation to drive user satisfaction, preventing fraud attempts, safeguarding users and driving revenues.

The head of data science at PhonePe, Kedar Swadi, told Analytics India Magazine that his team has a solid background in algorithms, mathematics and statistics. “It is always easy to learn new techniques once the basics are strong,” he added. 

Further, he said the team members have good programming skills and know-how to efficiently handle a large amount of data and create scalable pipelines. “We believe there is no compression algorithm for experience, and we specifically are staffed with people with deep domain expertise in financial services and, more narrowly, in payments,” said Swadi.

Team structure 

Currently, PhonePe has a 15-person data science team. “The team size grows based on the requirements we get from other teams in the organisation, and our plans for enhancing skills and capabilities in core data science as well as our expertise in the various domains that we work with,” said Swadi. 

PhonePe’s data science team consists of a mix of freshers and experienced professionals. The company follows a flat organisation structure.

“While all members focus on the core technical aspects, they also start contributing in other orthogonal areas such as data-driven innovations for business teams, mentoring junior members, process improvement for more predictable delivery, etc. as they grow,” said Swadi. 

PhonePe’s data science staff work in ‘pods’ (teams) to build long-term and short term solutions. 

Interview process 

Currently, PhonePe has three rounds of interviews: 

• A coding round where they test the candidate’s ability to use tools and libraries to analyse, interpret and visualise data
• A Q&A round, where they focus purely on the candidate’s understanding of core technical aspects (maths, statistics and algorithms) 
• An online (or offline) problem-solving session, where they give a problem statement and test the candidate’s ability to structure it into a crisp business problem, 

The last round of the interview is of paramount importance. Here, the interviewer looks at the candidate’s ability to distil the problem statement into data science tasks, explain the data required to solve the problem, describe and justify the techniques used to solve it, and then discuss how it can be presented to business or domain specialists.

Expectations

“There are three aspects that I see going into making a successful data scientist, viz., technical competence, business or domain expertise, and, very importantly, attitude,” said Swadi. 

“It’s worth mentioning that 80% of data science is hard manual work of looking at data, cleaning and processing it, understanding distributions in that data, and making sure you have a good understanding of that data,” added Swadi. It requires the humility to work on non-glamorous aspects, the tenacity to sift through vast amounts of data, the ability to deal with the failure of not producing positive results, and the strength to learn from it and get better.

Finally, PhonePe looks at competencies that describe the ‘how’ aspects of the job. “We look for intellectual curiosity, the ability to persevere even in the face of adversity, the ownership to get stuff done, and a penchant for setting and living a high bar in terms of standards,” said Swadi. 

Dos and don’ts

Swadi said, most candidates while interviewing for data science roles seem to be driven by the glamour factor, or the supposed monetary benefits. Also, a lot of candidates focus on the currently fashionable techniques (like assuming they only need to understand deep learning) rather than having a solid foundation in algorithms, maths, and statistics. 

Further, candidates tend to list techniques they don’t have a comprehensive understanding of (in terms of how they work, what assumptions they make about data, how they should evaluate, what impact tuning parameters has, etc.).

Other challenges, include: 

• Candidates do not have a sufficient understanding of the business aspects of problems in various domains they claim to have worked on. 
• Candidates often do not have sufficient exposure to the engineering aspects of developing solutions. Creating models is a small (even if important) part of the solution. 
• Candidates need to be more aware of issues such as productionizing, deploying, scaling, and closing feedback loop etc.

Work culture 

“One of the many great things about PhonePe is the culture,” said Swadi, “As a company, we are extremely data-driven. We thrive on taking up hard problems, taking calculated risks and experimenting when possible to arrive at better solutions.”  

PhonePe’s team gets full support from the leadership. “Support is given in terms of taking the time and explaining their problem, helping in extraction and understanding of their data, giving us the ability to experiment, and very importantly, accepting that we will sometimes fail,” said Swadi. 

Check out the data science job roles at PhonePe here.