Conquer the Data Science Interview – Part2
![Who doesn't like a person who talks data and business [Image by GraphicMama-team from Pixabay]](https://towardsdatascience.com/wp-content/uploads/2021/08/1mDyIfOklNL85a0pZl-hvxw.png)
DS interviews can be a saga of multiple rounds. Its common for most companies to conduct 2–3 rounds. This is sometimes followed by a director round/ hiring manager round. Multiple rounds means that you can be assured of being tested on all aspects of the DS project pipeline.
Broadly there are the following types of Questions:
1. Resume based/ Project based
2. ML Proficiency check – algorithm details
3. Metrics
4. Case-study based questions
We have covered type 1 and 2 in the previous post where we saw questions with a strong technical focus. The current post targets type 3 and 4.
![Case studies are a test of your thought process [Photo by Christina @ wocintechchat.com on Unsplash]](https://towardsdatascience.com/wp-content/uploads/2021/08/1QF3bNReXMzre5vZmFqq2nQ-scaled.jpeg)
Case-study based questions
In this type of question, the interviewer shares a Business use case, and asks you to solve it. Minimum guardrails are provided by the interviewer.
It is a way to check the candidate’s business acumen and see if they can take a vague business statement (yes, many a times the business requirements are indeed vague), ask relevant questions, state their assumptions, tie it down to a suitable Data Science solution, and finally translate the DS solution back into a language makes sense to business.
In such questions it is important to ask clarifying questions and break the problem into smaller chunks.
Eg1: Imagine you are an analyst at Big Basket. You have access to prices of all products in the BB catalogue. You also have access to transaction details. BB is planning to launch a premium honey developed in-house. You need to come up with a recommended price point for this new in-house premium honey. What will your approach be?
Concepts tested – Business acumen, data awareness
Eg2 : You are a part of the Quality control team at a company that manufactures vials for vaccines. The vials are supposed to have walls of minimum thickness of 2mm. The company guarantees a quality standard that has maximum defect rate of 3%. As a Quality tester how many vials do you need to test to be certain about whether the quality is met or not.
Concept tested – Statistics (Sample size estimation)
Eg3: You are a hired consultant for a boutique that makes various clothes and accessories. They sell their products through their own website. The boutique has different departments like – men’s clothing, men’s accessories, men’s shoes, and similarly structure for women. It is time for launch of the new spring-summer collection. However, due to low sales in the past few months there is an accumulated stock in the warehouse. Its your job to come up with a strategy that will empty the old stock and sell the new stock. You are allowed to give discount, but you need to be sensitive to each department’s cost.
State any assumptions you make. What data point will you consider to formulate your strategy? What kind of analysis will you perform on the data? Finally, how will you convince individual department heads to use your strategy?
Concepts tested: Business acumen, translate business to DS and vice versa
A possible solution could talk about the following points:
Data – Inventory management cost, cost of goods, past prices, past demands, customer transactions history
Analysis – Demand prediction, effect of offers, customer propensity to buy, product bundling
Communication with business – Show the graph of Sales Prediction without any strategy vs Sales prediction with proposed strategy. Likewise for Inventory costs with vs without any strategy.
Eg 4: You are a Data Scientist **** at a fintech startup. This company has a product that offers a line of Credit of $500 (unsecured lending). The target segment is people who do not have a regular income. Most banks insist on a credit check , pay slips, security/guarantee before providing a loan. But your company provides unsecured lending without nay credit check and doesn’t even charge interest. For anyone who approaches you, you gather their KYC details, and have access to last 24 months of transactions on their bank account. Come up with a strategy to identify if the bank account shared is their primary bank account, and whether they can qualify for the credit line.
Concept tested – Business acumen, ability to break the problem into chunks, awareness of the domain, translate business to DS
Eg5: You are a data scientist at a Digital Marketing firm. Your current project is on analyzing product reviews and identifying topics people are talking about. You have a high-volume incoming stream of data. How will you assign topics in real time? You need to be able to accommodate for new topics coming up in real time?
Possible solution could utilize word embedding.
![Is this the right measure? [Photo by Darling Arias on Unsplash]](https://towardsdatascience.com/wp-content/uploads/2021/08/1A4sQF4zXzg6Z9thy4ARmcw-scaled.jpeg)
Metrics (KPI)
"If you cant measure it, you cant improve it."
- Peter Drucker
For any business problem its important to know what you are trying to improve – reduce cost , increase engagement, increase profitability. ANd this is where metrics become important. I feel metrics are what tie the DS solution to the business objective.
Eg1: You are a Data scientist at Twitter. You want to increase the new user’s engagement with the platform. What metric will you use to figure out a new user about to become disengaged?
Hint – Track the trend of daily metrics – like logins and interactions.
Eg2: We are a mobile manufacturer. We have the historical price list of our mobiles and other companies’ mobiles We want to report to the management, how competitive our prices are with respect to the market. What will you do?
Hint – Proportions
Conclusion
As one progresses higher on the DS ladder from Junior to Senior Data Scientist the expectations of the job change, and the Interview Questions reflect that. Initially one is just expected to solve a problem provided the solution strategy is outlined. However, later one is expected to own a project end to end and that includes all aspects of the CRISP-DM methodology – defining the strategy, executing it, checking for its validity, deploying and monitoring the solution.
The diversity of questions mentioned in this series covers most of these aspects, and will help you prepare better.
You may also like: