For several years now, much of the discourse around emerging technologies has focused on one key thing: data science. This has spawned an information influx on how to professionally excel in this field.
However, the reality is that no matter what you majored in, or how many courses you juggled, or how well-prepared you think you may be, nothing can teach you better than your own experiences.
We spoke to five data scientists working across industries and vastly different scales to find out how their experiences inform the choices they make in their jobs, and the powerful lessons they draw from these episodes.
The importance of allowing business goals to guide your actions cannot be emphasised enough, as affirmed by most data scientists interviewed for this story. Scouring data sets and creating corresponding models does not yield solutions by itself. Before you conduct any analysis or even write a single line of code, justify its value by tying it to a business outcome.
“While data scientists are armed with unique tools and technologies to solve analytical problems, they should join forces with business functions to cash in on their acumen,” says Vidhya Veeraraghavan, Head of Analytics at Standard Chartered Global Business Services.
Concurs Navin Manaswi, Founder, CEO and Data Science Specialist at WoWExp Technologies. According to him, the key is to understand the business value of data science projects and mapping those onto solutions.
Be it acquiring customers or reducing costs, priority should be to quantify the potential impact of your project and align it to a direct business goal.
“The best and time-tested method that has given us the most desirable results is the ‘two-in-a-box’ model, where more people are jointly responsible for the effective implementation of a project,” says Veeraraghavan. “This gives an opportunity for both the data science team as well as the business functions to time-box their projects and work in parallel to obtain optimal solutions,” she adds.
While it is established that both teams should act as owners and contributors with the objective of failing fast and learning faster, AI and ML specialist at Publicis Sapient Sray Agarwal also feels that this approach helps build models that could inform the overall strategy of a business and hence, solve more than just one problem.
“To add to this, one of the biggest lessons I learned is the value of estimating the dollar figure for projects,” he says. “Until you can come out with a return on investment (RoI) and value realisation numbers, you will not be able to sell your idea to your clients,” he adds.
While it may be evident that ensuring where the data comes from and how it is going to be used is imperative before you commit to deadlines, take time to delve a little deeper at the beginning itself.
“Do not hesitate to ask uncomfortable questions during the requirement gathering phase,” says data scientist Usha Rengaraju, who is incidentally India’s first female Kaggle Grandmaster. “Spending a lot of time getting clarity on client requirements can save you a lot of headache towards the end,” she adds.
She, like other data scientists, sees the value in identifying stakeholders before beginning any data science projects and keeping them involved early on. Whether you are working with research analysts, marketers, or with the executive team, getting a good understanding of their workflow and the key challenges they face will help you mine data better.
“We regularly work with various stakeholders, each of whom have different perspectives and expectations,” adds Manaswi. “For instance, business leaders expect higher numbers in terms of RoI, as well as in terms of accuracy and robustness,” he adds.
While upskilling is a prerequisite for any profession, having an inquisitive nature and a growth mindset is very important as data scientists. Most follow a structured learning plan outside their workloads to build on their knowledge in areas such as AI, statistical techniques, and big data tech, among others.
“Strong programming skills can be a great asset and are vital, especially when solving problems in niche areas,” says Rengaraju. “The domains less travelled will not have many libraries or packages to quickly prototype ideas,” she adds.
To tackle this, take up courses online, listen to podcasts, watch tutorials, pick the brains of mentors, browse articles, practice coding, work on projects and like Rengaraju, participate in Kaggle competitions.
As data scientists, it is difficult to fight the urge to wander away from the problem statement and go down a deep rabbit hole. According to Data Scientist at RealPage Surya Prakash Manpur, with the plethora of tools and concepts coming up in this field, it becomes a challenge for them to not stray.
“Data science entails some complex methodologies, but all projects do not demand complex maths and algorithms,” says Manpur. “Some can be achieved with basic methods like correlation analysis and finding the significance of a variable. In fact, keeping the analysis simple and sticking to the problem statement is the biggest lesson I have learned,” he adds.
Rengaraju agrees. She feels that many times, big enterprise problems can be solved using simple statistical models. “Cutting-edge algorithms do not always win. The goal should be to use the best model to solve a given business problem.”
Storytelling techniques that help democratise analysis have been permeating the world of data science. Not only does it break down complex information that the data presents but also helps improve team cohesion.
“You need to weave a story around the problem statement and how your AI model can help generate creative solutions,” says Agarwal. Adds Manpur, “You should be able to convince your stakeholders about the decision you have made, or the approach you have taken to come up with your model.”
Data scientists have to work with a diverse team on a regular basis to employ analytics that aims to increase a business’ revenue or cut costs for the organisation. They are involved in every facet of the data lifecycle, which often means that they have to wear many hats.
But for all the datasets that they have worked on and practised in textbooks, the problems, in reality, are quite different, and expertise can come only with experience.