Record of my interview experience with QuantumBlack (McKinsey) as a data scientist in Singapore in 2019.
QuantumBlack (QB for short) may not be well-known to many people, but it was acquired by McKinsey in 2015, which made it a very reputable company. According to recruiters, QB’s employees are actually McKinsey employees and they share the same office in Singapore. QB’s data scientists are different from those in typical software companies as they are consultants specializing in data science who work on projects for clients.
You can find a lot of examples of personal experience interviews (PEI) and case interviews by searching “mckinsey interview” on Youtube. The official McKinsey website also provides interview examples, but they are mostly for general consultants and not for data scientists. Therefore, I am unable to find any resources specifically for data scientists, I provide one here.
Day 0: received message from recruiter on Linkedin
The recruiter sent a company introduction to me and asked if I was interested in the data scientist position at QB. I said yes and asked for the job description. Two days later, he replied to me with “Hi, xxx (not my name)” and a data engineer JD and asked if I was available from 6:30 to 8:30 the next evening. Was he confusing me with someone else? Later, I told him I was available, but he disappeared and never responded to me again. Later, I saw on Glassdoor that someone had the same experience, which is ridiculous! This kind of recruiter is not competent, right?
Day 7: received message from another recruiter on Linkedin
At first, the recruiter introduced the company and asked if I was interested in data scientist positions. Since it was from McKinsey, I gave the recruiter a second chance, and things went normally from there. After agreeing to proceed, I received an online assessment from Hackerrank, which consisted of three questions to be completed within two hours within seven days. The first question was a simple dynamic programming problem, roughly at the Leetcode easy level. The second question involved manipulating a Pandas DataFrame and was somewhat complex, requiring the use of functions that are not commonly used. I completed it in 45 minutes while looking up the documentation. The third question provided a dataset to train a model and return predictions. Even randomly answering this question would result in some level of accuracy, so I’m not sure what the passing criteria were.
Day 15: Feedback of Online Assessment
After submitting the online test, if you pass, the recruiter will contact you to schedule the next round interview. The recruiter congratulated me on passing the online test and mentioned that there would be no more exams. I thought to myself that the next steps would likely involve exams in areas where I’m not strong. In general, QB and McKinsey are the same company, and QB employees are considered full-time employees of McKinsey. Both teams work on projects together. This position requires frequent travel, and the Singapore office is responsible for McKinsey’s clients in Southeast Asia. For the next interview, I was given the option to either interview at the McKinsey office in Taipei or have an interview online. Of course, I wanted to visit the McKinsey office. I found out that the location is inside Taipei 101, which is really trendy.
Day 21: Onsite Interview
The interview location is on the 47th floor of Taipei 101, and access is required through a card reader at the entrance on the first floor. I hesitated whether to speak in Chinese or English, but ultimately used Chinese. I stated that I was there for an interview, and without asking for my name, the other party provided me with the access card. I wonder if they’re not afraid that someone might remember the wrong interview date and show up in person, or if everyone is opting for online interviews.
There were two interviewers, and I was taken aback when they appeared on the screen. I had been in the software company for too long and had forgotten that today was the management consultant interview. I wore a T-shirt casually and came over. Both of the interviewers were in suits without ties. However, they did not make any comments on my clothing.
One interviewer introduced themselves as a senior data scientist, while the other’s introduction was unclear due to my poor English listening skills. However, the second interviewer did not ask any questions throughout the interview and left halfway through to attend a meeting, so it seemed like it didn’t matter that I couldn’t understand their introduction. The first question was a PEI (Personal Experience Interview) asking me to introduce the ML projects I had worked on, including the features and models used. However, they didn’t seem very interested, and the question ended after three minutes. The main focus was probably similar to a typical consulting interview, asking about what I had accomplished in the project and what impact it had. Using the STAR (Situation, Task, Action, Result) method to answer would suffice. The key to answering the question is to tell a story that illustrates the impressive things I have done in the project.
The following is a case interview where the interviewer was conducting their first project at McKinsey. There are many examples of McKinsey case interviews available online, but I couldn’t find any examples specifically for data scientists, so I didn’t know how to prepare. Fortunately, the question was a typical machine learning question that would be asked in most machine learning job interviews, where you are given a problem and asked to solve it using machine learning. The interviewer asked in great detail, covering almost all aspects that should be considered in practical machine learning.
- What metrics to use
- What features can be used
- How to convert data of various types into fixed-length vectors to feed into the model
- Feature importance
- How to handle missing values under different circumstances
- What to do if test performance is poor
- Considerations for using linear models: handling missing values, feature scale, model assumptions
- Principles and differences between Random Forest and Gradient Boosting Decision Trees (GBDT), and what parameters to tune
- How to verify model performance offline
- How to do clustering
- How to determine if a set of random data follows a Gaussian distribution.
When I asked about the job responsibilities, I was told that it involves helping clients implement machine learning solutions or conducting data analysis, which could take anywhere from a few weeks to several months. It also involves writing code for clients. However, once the project is completed and handed over, if the client doesn’t have anyone who understands machine learning, they may need to pay McKinsey again for upgrades or revisions after six months or a year.
I originally guessed that there would be some behavioral questions, such as “What is the biggest setback you have encountered?” or “What is the most ambitious thing you have done?” However, interviewers did not ask those questions during the interview. The recruiter sent an email saying that they would give me feedback no later than the next day, but I still haven’t received anything yet. I will assume that I have been rejected.
Conclusion
The interview was tough, with detailed mathematical questions that weren’t too difficult, but when I got stuck, they gave me hints. They emphasized that there were no standard answers to most of the questions, which helped build my confidence. Most of the questions were asked in general machine learning job openings, so there was no need to prepare for anything unusual. They were particularly interested in model interpretability and wanted me to be familiar with how various models explain their predictions, which was easy to understand. Their clients were likely to be unfamiliar with machine learning, so explaining how the models work to them is also important. After all, explaining the model is also part of their job, but not every ML engineer needs this.