How the McKinsey data scientist interview works, based on six candidate reports: an online technical assessment, coding or debugging, ML and business cases, and Personal Experience Interview (PEI) style storytelling.
$143K
Avg. Base Comp
$165K
Avg. Total Comp
4-7 rounds
Typical Rounds
2 months
Process Length
The McKinsey data scientist interview combines technical screening with consulting-style cases, and candidates describe it as case-heavy. Several report a timed online assessment early on, such as a HackerRank test, and one also mentions a Quanthub questionnaire. Live coding or debugging can follow. One candidate had to fix code errors live in a HackerRank environment. Another was asked to narrate their reasoning while working through a dataset.
After the screens, candidates describe case discussions that blend modeling with business judgment. Examples include a time series forecasting case for a manufacturing plant, an injury-risk prediction scenario, and spotting anomalies in a claims dataset. One candidate said they went too technical when the interviewer wanted a data strategy answer. Another said the cases leaned on basic math and structured thinking more than advanced analytics. Plan to frame the problem and state your tradeoffs before you reach for a model.
Personal Experience Interview (PEI) style questions come up too. One candidate was asked to go deep on two or three projects, covering model selection, validation, and drift. They were then asked to re-explain each decision to a non-technical listener, for example how to justify higher compute costs. Another reached a final round on leadership stories and wished they had named the specific moment they changed people's minds. Reported formats varied by role, and none of the six candidates reported an offer, so treat the details above as individual experiences.
Synthesized from 6 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Mckinsey & Company process.
I interviewed with McKinsey and got eliminated because I could not solve the coding problem fast enough and I could not optimize it. I was able to get to a brute force solution, but I was not able to really optimize it. I do not remember the exact question, but it was LeetCode style.
They also asked me to talk about my experience, so it was not just pure coding. It was a mix of behavioral and a live coding interview, and the coding part is what I think knocked me out. I did not get to the later rounds.
The main thing I took away is that they wanted more than a working brute force answer. They wanted you to get to something cleaner and faster under time pressure.
Prep tip from this candidate
For McKinsey, do not stop at brute force. You need to get to the optimized version fast, because the live coding round seems to care a lot about speed plus improvement, not just correctness.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Mckinsey & Company
Get the top 3 highest employee salaries by department
| Question | |
|---|---|
| Month Over Month | |
| Over-Budget Projects | |
| P-value to a Layman | |
| Minimum Absolute Distance | |
| Categorize Sales | |
| Flipping 576 Times | |
| Resumable Fact Table Load | |
| Xgboost vs Random Forest | |
| Hurdles In Data Projects | |
| Production Model Monitoring | |
| Flatten JSON | |
| Binary Tree Validation | |
| Subway Machine Learning Model | |
| Stick Break | |
| User Event Data Pipeline | |
| Training vs Validation vs Test Data | |
| Model Deployment Preparation | |
| NxN Grid Traversal | |
| Deciding Between Solutions | |
| Client Solution Pushback | |
| Stakeholder Communication | |
| Why Do You Want to Work With Us | |
| Kindergarten Feasibility | |
| Underpricing Algorithm | |
| Your Strengths and Weaknesses | |
| Late Orders | |
| Statistically Significant Test | |
| PCA and K-Means | |
| Generative vs Discriminative |
Synthesized from candidate reports. Individual experiences may vary.
Candidates report an initial recruiter or HR screening call. One candidate was told the process would take about two months and that a HackerRank assessment came next. Another described the recruiter as communicative throughout. The exact content of this call may vary by role.
Candidates report a timed online assessment before the live rounds. Reported content includes SQL, Python and data manipulation, and data structures such as binary search trees. One candidate also described a Quanthub questionnaire covering regression, L1/L2 regularization and textbook-style ML theory. One internship candidate had three days to complete the assessment, and it took about an hour.
Some candidates report a live technical round. In one, a Principal Data Scientist asked fairly standard data science questions and had the candidate debug and fix code live. Another described a dataset-based round where they had to explain and defend their approach while working. A separate candidate was asked how to reduce overfitting in neural networks and to walk through algorithms from past projects.
Candidates describe case rounds in which the interviewer cares about structure and tradeoffs more than a single right answer. Reported scenarios include time series forecasting for a manufacturing plant, predicting football player injury risk, and anomalous claims patterns. One candidate said their cases used basic math and graph interpretation. Another was asked what data to add to improve a model.
Candidates report PEI or personal-experience sections with deep dives into past projects and leadership stories. One candidate was asked to explain each decision in plain language. Another described five evaluation axes and said they could ask which axis an interview would cover. Later rounds may include a partner or engagement manager. Outcomes reported were mostly no offer.
