Yes you read it correctly! I am calling my transition from being an MBA to being the Analytics Manager in a well known consumer retail brand a "WEIRD" one. And why do I say that? Because during my 5 year journey in data science, I have had the opportunity to work with a lot of business stakeholders like marketing head, brand managers, sales heads etc. and many a times they have asked me about my educational background. I would like to think that they asked this because of my ability to present the solutions keeping the business context and execution feasibility in mind. Well, the reason for asking this might be different for every individual, when I tell them that I am an MBA, their reply has always been the same, which is "What made you choose a technical career path after pursuing MBA?" And hence I decided to write this post to share my thoughts over 2 things:
1. Data science is not a completely technical field. Along with knowing Statistics, Algebra, Python, and the ever confusing "Confusion Matrix", it is equally important to understand business concepts like customer churn, campaign ROI, product bundling, merchandise mix etc. to be a successful professional in this domain
2. The other reason is I wanted to reach out to many non technical professionals who are from diverse educational backgrounds and share my learning path to make my way into this challenging yet super interesting career and how they can do it too.
Let's start with understanding The Business Side of Data Science with examples of few real life projects:
1. Customer Segmentation: This project was assigned to me by the marketing head of one of the business divisions and the objective was to bucket the 10 million customer base into 5 to 6 segments where each segment defines the profile of consumers lying in that segment. For example one segments can be customers who seek discount and make a purchase only during offer periods, another segment can be consumers who have an affinity towards purchasing fashionable and high value products etc. Now when I speak to and interview the aspiring data science professionals, they know about various clustering techniques like K-means and Hierarchical clustering but have very limited understanding of how to approach this kind of a project in a real world scenario. Trust me, running these algorithms constitutes a small part to the success of such projects. The key is to define:
- What variables will you use to segment customers? Eg. Average discount%, average bill value, special occasion buyers, age group, loyalty with the brand etc. These are few variables which I used during the data preparation stage and this is where my MBA knowledge gives me an edge
- How will the business use these segments? Will the segments be used to target customers with personalized product offerings or is it to understand their buying behavior? Is this intended to be only a demographic based segmentation or should I include their transaction profiles as well? Again, these are the questions which you as a data scientist would need to ask the business person. They will generally give you a very broad objective and if you don't ask such questions upfront, the project can go through a lot of unnecessary iterations.
Once I got my answer to the above mentioned questions, I was able to do the data preparation with ease and also profile the segments clearly which were ready to be consumed by the marketing team.
2. Customer Churn Propensity: This was another key project where my business acumen helped me deliver an effective output. The business question in this case was to identify customers who are likely to churn. Here again asking few critical questions was the key:
- At what level do we want to define churn propensity? Since our business sells multiple product categories, it was important to ask if we want to define customer churn at a category level or at overall business level. Both these contexts are completely different as if we go at category level then we would have multiple models with churn propensity for each category and in other case we will have only a single model. Also if we go at category level, we would need to study consumer purchase behavior for each category. For example, on an average customers might take 3 months to repeat for category like a perfume but 9 months to come back to buy sunglasses.
- What is the objective of this exercise? Do we want to start a program to prevent customer churn or do we want to filter out such customers from our campaigns altogether.
Asking such questions helped me prepare my hypotheses better along with coming up with a right approach for exploratory data analysis.
I hope I have presented my case well for the importance of having good business acumen to be an effective data scientist. However this doesn't mean that you necessarily need to have an MBA degree. The point that I would to bring to your notice is that having a decent understanding of how businesses work is a key skill that every data scientist must work on.
The other piece that I wanted to share is how non technical professionals can find a career in data science:
1. Find a mentor: Since you are coming from a non technical background and have developed interest in data science, it is important to get advise from someone who has at least 5 years of experience in the industry. Talking to a mentor will give you a lot of clarity about the kind of roles in data science, the kind of things these people do on a day to basis, which technologies they use and what are the hard as well as soft skills that you need to develop yourself as a successful data scientist. And you can easily find a mentor through your own personal network or if you don't know anyone, then you can reach out to data science professionals through Linkedin. Trust me, at least 90% of the people would be more then happy to help.
2. Upskill through an Online Course: Being a newbie in this field can get very confusing many a times as you won't be able to understand where to start and which all topics and concepts to learn first. You must understand that this is a highly knowledge intensive field and even professionals with years of experience have to keep learning new things on a daily basis to stay relevant as their are new techniques coming out very frequently. Hence it is very important to have a set learning path and there are many well structured courses which you can find online. I am listing out a few for your reference. However this is not a comprehensive list and I would urge you to do your own research before enrolling for any course:
- Data Science Career Track by Springboard
- Machine Learning Course by Stanford (Taught by Andrew Ng)
- Get a DataCamp Subscription
- Machine Learning A-Z by Udemy
3. Participate in Data Science Competitions: Data science competitions on portals like Kaggle and analytics vidhya can give any newbie an opportunity to work on a wide variety of data science problems and are by far the best source to gain experience on real world data science projects. I would urge you to start with the below mentioned popular competitions first as you can find enough help for these projects as thousands of data scientists have already participated in these projects and have shared a lot of good solutions and approaches to solve these problems. The idea here is not to copy these solutions blindly, but since you are starting, it will be helpful to get a hand holding in initial projects. After you have worked on 3 to 5 such projects, I am sure you would gain the confidence to solve these problems independently. Below I am mentioning a list of popular data science competitions and also a few articles with there solutions:
A. Titanic: Machine Learning From Disaster: This is actually a legendary machine learning competition which almost every data scientist has dabbled with. You will get a deep understanding of how to approach and solve classification problems. Below are some good solution links:
B. Digit Recognizer Challenge: This is a slightly advanced challenge but will definitely increase your interest further in this amazing domain of data science as you will get a hands on understanding of computer vision. Below are the solution links:
- My Journey to Top 10 of Kaggle Digit Recognition Challenge
- Digit Recognizer - Introduction to Kaggle Competitions
C. Predicting House Prices: This is a regression problem and again a very popular data science competition. Below are solution links:
The above 3 projects cover a mix of classification, regression and computer vision problems and are a very good starter for learning data science.
Apart from the above 3 points, i.e. finding mentor, enrolling in a structured course and participating in data science competitions, I would advise you to ensure that you to hands on practice on a daily basis.
I hope through this article I was able to give all the non technical professionals a sense of confidence about why they have the inherent ability to make a good data scientist and also some motivation to take up their data science journey. Happy Learning and keep going!!