← Back to Directory

Capital One interview notes, page 35 of 39

Large Enterprises

Most of these Capital One write-ups come from Data Scientists (196), then Software Engineers (93) and Product Managers (54). The most reported rounds are the technical phone screen (213), the product sense round (58), the behavioral / leadership round (41) and the system design round (22).

382 interview notes · updated Oct 2026

Capital One·Product Manager·Onsite - Product Sense / Strategy

Jul 2026
Capital One PM interview with a product strategy question about growing their shopping product. Pretty open-ended, which I wasn't fully ready for.
  • How would you recommend a growth strategy for Capital One Shopping?

“I went straight into user acquisition channels and kind of forgot to anchor on what 'growth' even meant for this product specifically.”

View Post

Capital One·Data Scientist·Technical Phone Screen

Jul 2026
Capital One data scientist interview that went deep into subscription unit economics and insurance math. Six parts back to back, some straightforward, some genuinely tricky once churn and fixed costs entered the picture. Left feeling okay about the first half and less sure about the back half.
  • A subscription service charges $40/month with the first 3 months free, a $25/month variable service cost, $35 one-time installation, and $120 variable marketing/overhead per customer. The contract is 15 months with no churn. What is the net value per new customer?
  • Using the same subscription parameters but extending the contract to 18 months, recompute the net value per customer and explain exactly why the result changes.
  • Now extend the contract to 21 months with 10% of customers breaking the contract uniformly across paid months and paying a $100 penalty. Marketing drops to $20/customer variable, overhead becomes $1,000,000/year fixed, installation stays $35. For N customers acquired, derive a formula for the minimum N to break even company-wide.
  • Qualitatively describe the demand vs. price curve and the total profit vs. price curve for this subscription service. Where is revenue maximized versus where profit is maximized, and why can those two points differ?
  • An insurer sells a 12-month policy at $30/month premium. Servicing costs $3/month. A covered claim pays $8,000 in benefits plus $300 in regulatory costs. There's also a $4/quarter regulatory expense regardless of claims. Let p be the probability of at least one claim in the year. Write and solve the equation where expected profit equals zero.
  • Given 400 policies observed over a year with 18 claims, compute the MLE for the claim probability, build a 95% Wald confidence interval, explain when the Wald interval breaks down, and suggest a better alternative.

“Pretty mechanical once you lay it out.”

View Post

Capital One·Software Engineer·Onsite - Behavioral / Leadership

Jul 2026
Behavioral round for a Data Engineer role at Capital One, all three questions were some flavor of 'tell me about a hard time' with a follow-up on what you'd do differently. Nothing too surprising but you really need concrete stories ready or it gets uncomfortable fast.
  • Tell me about a time you had a conflict with a cross-functional partner like a PM, engineer, or someone from legal, and how you handled it.
  • Describe a time you ran into a serious technical problem, whether it was a modeling issue, a data quality problem, or a production incident, and walk me through how you resolved it.
  • Can you give an example of a time you had to work through unclear or shifting requirements and still managed to deliver something meaningful?

“This one tripped me up because my first instinct was to pick a story where I 'won' the disagreement, which is the wrong move.”

View Post

Capital One·Data Scientist·Technical Phone Screen

Jul 2026
Capital One data scientist interview with a pretty involved business case around Groupon partnership economics. One question but it had a lot of moving parts and felt more like a strategy consulting prompt than a typical DS interview.
  • You're a restaurant owner. Given specific unit economics (variable costs at 40% of spend, fixed costs $100/day, vouchers worth $30 sold at $15 with a 40% commission taken), decide whether to partner with Groupon. Walk through your decision criteria, key risks and how you'd mitigate them, and give a concise exec-ready recommendation including the assumptions that need to hold and what would make you pause or kill the program.

“This one took me a minute to even figure out where to start.”

View Post

Capital One·Data Scientist·Technical Phone Screen

Jul 2026
Capital One data scientist interview with a heavy quant/probability focus. The whole session was basically three variations of an optimal stopping problem with a die, plus a generalization at the end. Not what I expected walking in.
  • You can roll a fair six-sided die up to three times and stop whenever you want, taking that roll's value in dollars. If you haven't stopped by the third roll, you must take it. What's the optimal stopping strategy and the expected payout?
  • Same three-roll die game, but now each time you choose to continue after a roll, you pay $1 before the next roll. What are the optimal stopping thresholds and expected payoff under this cost structure?
  • Now in the same three-roll setup, you always pay $1 to access each subsequent roll regardless of when you decide to continue. So roll 2 costs $1, roll 3 costs another $1. Find the optimal policy and expected payoff.
  • Generalize the second game (with continuation costs) to an n-sided die with faces 1 through n and a continuation cost of c per roll. Derive the stopping thresholds as a function of n and c.

“This is the classic secretary-ish stopping problem and I'd seen something like it before, so I wasn't totally lost.”

View Post

Capital One·Software Engineer·Technical Phone Screen

Jul 2026
Capital One software engineer interview with two meaty coding problems back to back. The matrix manipulation part was fine but the street lights problem took a while to click and I'm not sure I nailed the complexity analysis on the spot.
  • Given an m×n integer matrix, implement several in-place operations: swapping two rows, swapping two columns, reversing row order, reversing column order, and rotating the matrix 90 degrees in either direction. For the rotation, handle both square and rectangular matrices and explain your time and space complexity for each operation.
  • You have a list of street lights, each defined by a position and a radius, illuminating a segment on a 1D line. Find the point covered by the most lights. If there's a tie, return the smallest such coordinate. Implement an optimal solution and analyze time and space complexity.

“The swap and reverse stuff was pretty mechanical.”

View Post

Capital One·Data Scientist·Onsite - Product Sense / Strategy

Jul 2026
Capital One data scientist interview with a product strategy angle, basically asking you to think like a PM about credit card benefit abuse. One question, but it had some teeth to it.
  • Some cardholders take advantage of location-based perks without ever using the card for actual purchases. What would you propose to reduce this kind of behavior and bring spending back in line with benefit usage?

“I went straight to minimum spend thresholds and felt pretty good about it, but then kind of stalled when trying to get specific.”

View Post

Capital One·Data Scientist·Technical Phone Screen

Jul 2026
Capital One data scientist interview that was basically a deep stats exam disguised as an A/B testing question. One long multi-part problem covering everything from t-test mechanics to effect sizes to robustness arguments. Dense.
  • Given two independent samples from an A/B test on per-user daily watch time, compute both the pooled-variance t-statistic and the Welch t-statistic, then report the corresponding p-values for each.
  • The product goal is to increase watch time. Should you use a one-tailed or two-tailed test, and why? Justify your choice rigorously.
  • Construct a 95% confidence interval for the mean difference in watch time using Welch's method.
  • Compute Cohen's d and Hedges' g for the effect size between the two groups.
  • If the normality assumption is questionable but sample sizes are large, why is the t-test still valid? And if heavy tails are suspected, what robust alternative would you propose and what are the trade-offs?

“I knew the formulas but fumbled the pooled variance denominator at first, mixed up whether you use nA+nB-2 or something else.”

View Post

Capital One·Data Scientist·Technical Phone Screen

Jul 2026
Capital One Data Scientist interview with a product profitability case built around a cloud startup's $54 premium plan. It's more analytical than you'd expect for a DS role, closer to a business case than a modeling question.
  • A cloud startup charges $54/month for a premium plan. What factors drive monthly profit for that plan, how would you write out a profit equation, and what data would you need to actually compute it? If the interviewer gives you a missing ratio like free-to-premium conversion rate, walk through the full calculation and say what the number means.

“I started listing revenue and costs which felt right, but I kept second-guessing whether to include acquisition costs as a fixed or variable item.”

View Post

Capital One·Data Scientist·Onsite - System Design / Architecture

Jul 2026
Capital One DS interview that was basically a full system design gauntlet for a real-time fraud detection pipeline. Seven parts covering everything from label leakage to adversarial fraudsters. Walked out feeling like I'd just defended a thesis.
  • How would you construct time-aware train/validation/test splits for a fraud model where labels like chargebacks arrive up to 14 days after the transaction?
  • Propose ten or more features for a card fraud model and explain how you'd handle high-cardinality categoricals and avoid target leakage.
  • Compare supervised models like XGBoost versus anomaly detection approaches like Isolation Forest for fraud detection with very sparse positives. When would you combine them?
  • Given a cost matrix where false positives cost $5 and false negatives cost $200, how would you set a classification threshold and what metric would you optimize during model selection?
  • How would you monitor a deployed fraud model for drift and degradation when ground truth labels don't arrive until 14 days later?
  • Design a safe rollout plan for a new fraud model, including shadow mode, holdback groups, and guardrails to limit business risk during the transition.
  • Describe three defenses against adaptive fraudsters who learn to game your model over time, and explain how you'd validate that those defenses actually work.

“This one I actually felt decent about.”

View Post