Capital One Interview Questions — Page 16

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

May 2026

Capital One data scientist case study round, focused on a marketing analytics scenario about waiving annual fees. Pretty heavy on experimentation design, which I should've seen coming given the role.

You're evaluating a marketing campaign that waives the first-year annual fee for new cardholders. Design an A/B test to measure its impact on card acquisitions and spending. Walk through your primary metrics, guardrail metrics, how you'd size the experiment, and what post-experiment analysis would look like.

“I started with conversion rate as the primary metric and revenue lift as a secondary, which felt right.”

View Post

Capital One·Machine Learning Engineer·Onsite - Behavioral / Leadership

May 2026

Behavioral round for an ML engineer role at Capital One. Just one question from what I can tell, but it was the kind that takes the whole session if you let it.

Tell me about a time you had to work with a difficult teammate in a research or technical setting.

“I went with a story about a collaborator who kept pushing back on my modeling choices without much technical justification.”

View Post

Capital One·Data Scientist·Technical Phone Screen

May 2026

Capital One data scientist interview that went deep into unit economics and pricing models for a cloud storage case. More math-heavy than I expected for a DS role, felt like a finance case study at points.

Given a cloud storage startup with 20 subscribers, $400 fixed monthly cost, $5 variable cost per user, and $1/GB pricing with average usage of 15 GB per user per month, compute monthly revenue, total variable cost, total cost, and profit.
Explain numerically why the firm is operating at a loss, and identify which lever (price, usage, user count, fixed cost, or variable cost per user) has the most impact on profit in this scenario.
If the subscriber count doubles to 40 with the same usage distribution, does the firm break even? If not, what exact subscriber count is required for break-even?
A two-tier pricing model is proposed: free users generate $0 revenue and cost $1/month each, paid users generate $54 revenue and cost $5/month each. Fixed cost stays at $400. With 20 total users, what minimum percentage must be paid tier to break even? Also derive the general formula for minimum paid share as a function of total users T.
If only 25% of users convert to the paid tier, how many total users are needed to break even under the two-tier model? Set up the inequality and find the smallest integer solution.

“Straightforward arithmetic but I second-guessed myself on whether variable cost was per-user or per-GB.”

View Post

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

May 2026

Capital One data scientist interview that went deep fast. One system design question that touched streaming pipelines, probabilistic data structures, and complexity guarantees all at once. Not a vibe check.

You're receiving an unbounded stream of events, each with an event time, user ID, and category. Events can arrive up to 48 hours late and are out of order. Design an API with an ingest method and a query method that returns, for each hour in a given window, the count of unique users and the 95th percentile of per-user interarrival delay based on event time. Late arrivals must retroactively correct prior aggregates, ingest must be amortized O(log n), memory must be bounded relative to the correction window and active categories, and queries over H hours must run in O(H + C·log W). Walk through your data structures, how you handle clock skew and deduplication, and write test cases that cover late data arriving right at the 48-hour boundary.

“This was a lot to hold in your head at once.”

View Post

Capital One·Data Scientist·Technical Phone Screen

May 2026

Capital One data scientist interview that went deep into ML system design for donor solicitation optimization. Five meaty questions back to back, all connected to the same business scenario. It felt more like a case study session than a traditional technical screen.

Design a model to predict expected net revenue from donor solicitations. Walk through whether you'd use a two-stage approach (predict conversion and amount separately) or model expected revenue directly, and cover feature choices, sparse/categorical handling, leakage risks, and how you'd deal with heavily skewed donation amounts.
Given your model's predicted probabilities and amounts, how do you define a profit-optimized decision policy for the online channel and for gala invitations? The gala has a fixed venue cost of $20,000, capacity of 100, and costs $100 per attendee. Show how you'd rank and select donors.
Why does calibration matter specifically for profit optimization? Pick a calibration method and describe how you'd evaluate it.
How would you evaluate this policy offline when historical solicitations weren't randomly assigned? Describe inverse propensity weighting or doubly robust estimation and what assumptions each requires.
If you reframe the online campaign as an uplift modeling problem instead of a pure prediction problem, what's the objective, what metrics would you use to evaluate it, and how do you prevent the model from systematically ignoring lower-value donor segments?

“The skewness piece is where I spent most of my time and I think it paid off.”

View Post

Capital One·Data Scientist·Technical Phone Screen

May 2026

A Capital One Data Scientist interview that was essentially a finance math problem dressed up as a case question. One question, pretty self-contained, but easy to fumble if you rush the arithmetic.

Given fixed costs of $375M, sales of 231M units at $4 each with a $1 unit cost, calculate the Year 1 profit margin (profit divided by revenue) as a percentage to two decimal places, walking through each step.

“The math itself isn't hard but I almost tripped up by forgetting to subtract fixed costs before dividing.”

View Post

Capital One·Machine Learning Engineer·Technical Phone Screen

May 2026

Got a coding screen for an ML Engineer role at Capital One. Pretty straightforward string manipulation problem, nothing that screams machine learning to me but whatever.

Given a string, scan every window of three consecutive characters and count how many times the first and last characters in that window are the same letter (case-insensitive). Return the total count.

“Not hard but I second-guessed the case-insensitive part for a second and almost forgot to lowercase before comparing.”

View Post

Capital One·Data Scientist·Technical Phone Screen

May 2026

Capital One data scientist interview with a pretty involved restaurant economics case. The whole thing was basically one long quantitative problem broken into parts, and they wanted you to show your work and label everything precisely.

Given a restaurant with 20 tables per day, $30 average revenue per table, variable costs at 40% of revenue, and $100 fixed costs daily, what is the baseline daily profit?
If all 20 tables used a coupon and the coupon platform takes a 40% commission on each table's check, what average check per table would keep daily profit equal to the baseline?
After joining the coupon platform, you now see 25 tables per day, 10 of which used a coupon. The overall average check across all 25 tables is $36. Variable and fixed costs are unchanged. Calculate the new daily profit and decompose the change versus baseline into three parts: the effect of more tables, the effect of a higher average check, and the cost of commissions on coupon tables.
Based on the numbers from the previous scenario, is profit up or down versus baseline, by how much, and what is the main driver of that change?

“Straightforward setup question but I second-guessed myself on whether VC applied to gross revenue or net.”

View Post

Capital One·Data Scientist·Technical Phone Screen

May 2026

Capital One data scientist tech round centered on unit testing an existing class. Three-part question that escalated from reading comprehension to writing actual pytest code, which is not what I expected from a DS interview.

Looking at the provided unit test, what does it actually validate? Walk through what the test is checking.
What additional test scenarios would you add to this test suite, and why would each one matter?
Pick the single most critical missing test scenario and write the actual pytest code for it.

“Easier than it sounds but also kind of a trap for overconfidence.”

View Post

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

May 2026

PM interview at Capital One with a product analytics case about Instagram engagement. Just the one question from what I can tell, but it had some depth to it.

You're a PM at Meta. DAU for Instagram is down while MAU is holding steady. How do you diagnose and address this?

“This tripped me up more than I expected.”

View Post