Interview guides · Hugging Face

Hugging Face Data Scientist Interview Questions & Guide 2026

Practice explaining analysis before a Hugging Face data scientist interview: the recommendation, the uncertainty, the follow-ups.

The 60-second brief

As a Data Scientist at Hugging Face, you play a pivotal role in shaping the future of machine learning and natural language processing technologies. This position is critical not only for the development of groundbreaking AI models but also for enhancing user experiences across various applications. By leveraging large datasets and advanced algorithms, you will contribute to products that empower developers and organizations to integrate AI seamlessly into their workflows.

How much statistics you need depends on the flavour of the seat. Experiment-facing work wants you deep enough to notice that repeated looks at accumulating data inflate the false positive rate of a fixed-sample test; modelling-facing work wants estimation and honest uncertainty intervals.

Focus areas

Strip CI, retry and synthetic traffic first
Read NRR on a fixed account cohort
Analyse at the account grain, cluster errors

Overview

The impact of your work as a Data Scientist extends to real-world applications, such as improving conversational agents, optimizing recommendation systems, and enhancing sentiment analysis tools. You will collaborate with cross-functional teams, including engineers and product managers, to drive innovation and ensure that the AI solutions developed meet user needs and align with the strategic goals of Hugging Face. This role offers a unique opportunity to work at the intersection of technology and creativity, making significant contributions to projects that are at the forefront of the AI revolution.

The Hugging Face interview process

Hugging Face candidates report 4 rounds · ≈ 3-5 weeks. The stages below are what candidates describe, not a published process.

Introductory Conversation

An added round often puts you in front of someone outside the core hiring team: a partner engineer, a product owner, a domain expert, sometimes a more senior manager. The question they are really asking is not whether you can do the work but whether they would trust a number that came from you. That changes what a good answer looks like. Lead with what the decision cost and what it changed, keep the method available but not central, and be plain about the limits of your evidence. Overstating a result is the fastest way to lose this round.

What to demonstrate

Whether you can explain a technical choice to someone who will never read your code, without either flattening it into nothing or hiding inside jargon
Honesty about evidence strength: what the analysis establishes, what it only suggests, and what it cannot say at all
How you take disagreement, specifically whether you update on a good objection, hold your position with reasons, or fold on contact

How to prepare

Write the two-sentence version of your most technical project for a non-specialist, then check that neither sentence needs a method name to make sense.
For one result you are proud of, write the strongest objection someone could raise and a response that concedes the part of it that is correct.
Prepare one decision that turned out to be wrong: how you found out, what it cost, and what you changed afterwards. A senior cross-functional interviewer asks for this more often than a technical one does.

Technical Assessment

This round decides whether someone can hand you a schema and a question and trust the number that comes back. Correctness under a clock is the bar, not clever syntax. The habit that separates strong from weak answers is checking the grain: after every join, know how many rows you expect and whether the count moved. Most wrong answers in this format are not wrong logic, they are a fan-out from a key that turned out not to be unique, or a filter applied before an aggregate when it belonged after. Say what you expect before you run it.

What to demonstrate

Whether your row counts survive each join, and whether you notice on your own when they do not
Deliberate handling of rows that fail to match, including whether the question needs an inner join or a left join with the non-matches kept and counted
Whether NULLs are treated on purpose, given that a NULL compares equal to nothing and that COUNT of a column skips it
Reaching a defensible answer inside the window instead of a refined one after it

How to prepare

Take a two-table schema, write a join that fans out on purpose, then fix it by collapsing the many-side to one row per key before joining. Repeat until the fix is reflex rather than recall.
Write a funnel as one query and print the distinct user count at each stage, then confirm each stage is a subset of the one above it rather than assuming it
Do a few timed runs in a plain text box with no autocomplete and no formatter, since assessment editors often have neither

Technical Interviews

A handful of shapes account for most of what gets asked in this format: a ranking or deduplication inside groups, a running or rolling total, a period-over-period comparison, and a cohort tracked forward over time. Recognising the shape quickly is most of the speed here; deriving it from scratch while a clock runs is where the time goes. Know that a window function keeps every row while a GROUP BY collapses them, and know which one the question needs. If the exercise is in Python instead of SQL, the same shapes arrive as groupby with transform, shift and merge, and the same grain mistakes are available.

What to demonstrate

Whether you reach the right construct without a detour, such as ROW_NUMBER over a partition to deduplicate instead of a self-join against a MAX subquery
Whether you know what your window frame actually is, since adding ORDER BY inside OVER changes the default frame and silently changes a running total
Whether the thing runs. A near-miss that throws an error scores below a plainer query that returns the right rows.

How to prepare

Write each of the four shapes once from memory against a small schema and keep the working version somewhere you will reread it: dedupe with ROW_NUMBER, a running total, a month-over-month change with LAG, and a retention table
Compute one running total twice on data with tied timestamps, once on the default frame and once with ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW, and look at where the two disagree
If Python is on the table, rebuild the dedupe and the running total with groupby and cumsum, then assert the two implementations return identical rows

Cultural Fit Assessment

Before anything else, this round is a reading test. You are given a small schema and a question phrased in business language, and most of the difficulty sits in the gap between them. Who counts as an active user, does a refunded order still count as an order, is that date column an event time or a load time. Weak answers start typing immediately and compute something precise about the wrong population. Strong ones pin the definition in one sentence, name the column that encodes it, then write the query. On a timed assessment with nobody to tell, write the definition in a comment anyway.

What to demonstrate

Whether an ambiguous term becomes a specific column and filter before any computation happens
Whether you read the schema for keys and cardinality rather than only for column names
Whether the result answers the question at the grain it was asked at, per user or per session or per day

How to prepare

Take three metrics you already use and write down the exact filter and exact grain behind each, then practise stating one of them in a single sentence out loud
On a schema you have never seen, spend the first minute writing what one row of each table means and which key it is unique on, then predict which joins can duplicate rows
Rehearse a version where the definition changes halfway through, and edit the query you have instead of starting over

Where candidates lose points

Reading consumption metrics before the metering lag window has closed

Usage pipelines land late and correct themselves, which is exactly what is_restated and restated_at record. A dashboard queried on day T sees a partially populated tail for the last several days, so the most recent points always slope downward and always look like a regression. Analysts then explain the artefact, and sometimes ship a change to fix it. Establish the empirical settling time by measuring how much a given usage_date's total moves between first_written_at and its final value, exclude that many trailing days from every reportable figure, and never compare a fresh period against a settled one.

Computing monthly churn against the entire customer base when contracts are annual

An annual contract has no opportunity to churn except at its renewal date, so an account that is eleven months from renewal is in the denominator while being incapable of appearing in the numerator. The resulting rate is smaller than the real one by roughly the ratio of the base to the renewal-eligible base, and it oscillates with the seasonality of when deals were originally signed rather than with anything about the customers. The corresponding trap on the other side is counting a churn on the date the record was updated rather than on term_end_date, which shifts losses into whichever month the operations team did its paperwork.

Reporting a p-value with no effect size or interval

Give the estimated difference with a confidence interval in the units the business cares about, then say whether that whole interval is worth acting on. A p-value only addresses whether you can rule out exactly zero; it says nothing about magnitude.

Over-explaining the method and under-explaining the implication

Lead with the answer and what you would do about it, then give the approach when asked. Roughly one sentence of method per three of implication is the right ratio for a stakeholder-facing answer; the interviewer already knows what a regression is.

Questions & practice

Explain a scenario in which you would use a decision tree over a random forest.

Medium · machine learning and modelling

Approach

Frame the prediction: the label, the moment of prediction, and the action it triggers.
Say how the offline result would be validated online before it is trusted.
Pick an evaluation metric that matches the cost of each error type, not a default.

Follow-up

How would you choose the decision threshold, and who owns that choice?
Where could label leakage enter this setup?

How would you optimize a logistic regression model?

Medium · machine learning and modelling

Approach

Check what information would not exist at prediction time, and exclude it.
Frame the prediction: the label, the moment of prediction, and the action it triggers.
Say how the offline result would be validated online before it is trusted.

Follow-up

How would you choose the decision threshold, and who owns that choice?
What would you monitor after launch to know the model is still valid?

Quantify billable volume created by retries after server errors

Hard · telemetry-hygiene · pandas · unit-economics

From fct_api_request (request_id, account_id, endpoint, idempotency_key, is_retry, http_status, request_at, billable_units), measure how much billable volume in a 28-day window is retry traffic that followed a 5xx. Group requests into attempt chains by (account_id, endpoint, idempotency_key) ordered by request_at; a request is error-driven if any earlier attempt in its chain returned 5xx. Requests with a null idempotency_key cannot be chained, so report them as their own class rather than assuming each is unique. Return billable_units split into first-attempt, error-driven retry, other retry and unchainable, per account.

Approach

Split the population before measuring anything. A null idempotency_key is not a chain of one, it is an unknown; report its share of billable_units first, because if it is 40% of volume then the headline estimate is a lower bound and the deliverable has to say so.
Within chainable rows, sort by (account_id, endpoint, idempotency_key, request_at) and derive 'any earlier attempt failed' with arithmetic rather than a per-group lambda: with is5 = (http_status >= 500), the per-chain cumsum minus the row's own value is positive exactly when an earlier attempt in that chain returned 5xx. A groupby-apply gives the same answer and is unusable at five million rows.
Do not take is_retry as the definition. It is set by the client whenever an idempotency_key is resent, which covers retries after client-side timeouts and after 4xx as well; compute the flag yourself and then cross-tabulate it against is_retry, because the disagreement is a finding in its own right.
Aggregate billable_units by (account_id, class) and assert the classes sum to each account's total. The spine sets billable_units to zero on 5xx responses, so the failed attempt contributes nothing and the whole inflation sits in the successful retry that follows it.
Report the per-account share and look at its distribution, not the fleet total. One account in a retry storm dominates any blended figure, which is the same failure that makes a fleet-wide error rate useless.

Worked solution (40 min)

unchainable = df.idempotency_key.isna(); report df.loc[unchainable].groupby('account_id').billable_units.sum() before proceeding.
keys = ['account_id','endpoint','idempotency_key']; c = df[~unchainable].sort_values(keys + ['request_at'], kind='mergesort'); c['is5'] = (c.http_status >= 500).astype(int)
c['attempt_no'] = c.groupby(keys, sort=False).cumcount(); c['prior_5xx'] = (c.groupby(keys, sort=False).is5.cumsum() - c.is5) > 0
c['cls'] = np.where(c.attempt_no == 0, 'first_attempt', np.where(c.prior_5xx, 'error_driven_retry', 'other_retry')); out = pd.concat([c, df[unchainable].assign(cls='unchainable')]).groupby(['account_id','cls']).billable_units.sum().unstack(fill_value=0)

Expected result: A per-account table of four columns summing exactly to that account's total billable_units in the window. error_driven_retry is zero for any account with no 5xx; the fleet-level error-driven share is typically low single digits while a small number of accounts sit far above it.

Verify it yourself

out.sum(axis=1) equals df.groupby('account_id').billable_units.sum() for every account.
Every row classed error_driven_retry has attempt_no >= 1.
An account with no http_status >= 500 in the window has error_driven_retry == 0.
Cross-tabulating cls against is_retry: no first_attempt row should carry is_retry True, and any that do mark chains truncated by the window boundary.

Follow-up

An account's error-driven share is 22%. Is that the platform's fault or the client's, and what do you look at next?
How would you define a consumption-based north-star metric that an outage cannot inflate?
Chains straddle the 28-day boundary. How large is that bias and in which direction?

Solve a coding challenge involving data manipulation using Python or R.

Medium · sql and data manipulation

Approach

Compute rates by summing numerator and denominator separately, never by averaging rates.
Check whether any join is one-to-many before aggregating, or the sums inflate.
Handle the rows that do not match: a LEFT JOIN with a NULL check is usually the question.

Follow-up

What breaks if events arrive late or out of order?
How would you verify this result without re-running the same query?

Collapse retried requests into logical operations per account

Medium · window functions · deduplication · null semantics

fct_api_request carries request_id, account_id, api_key_id, idempotency_key (null when the caller supplied none), is_retry, request_at and http_status. For one ISO week, count logical operations rather than HTTP calls: requests sharing (account_id, api_key_id, idempotency_key) collapse to the earliest one, while every request with a null idempotency_key is its own operation. Return per account the raw request count, the logical operation count and the ratio between them, and state which direction a rising 5xx rate pushes that ratio.

Approach

Split the stream before deduplicating. GROUP BY and PARTITION BY treat NULLs as equal to one another, while the = operator does not, so a single grouping pass folds every null-key request in an account into one 'operation' and can cut the count by orders of magnitude.
Deduplicate the keyed rows with ROW_NUMBER() OVER (PARTITION BY account_id, api_key_id, idempotency_key ORDER BY request_at, request_id) and keep rn = 1. Including request_id in the ORDER BY makes the result deterministic when two rows share a millisecond, which they will.
UNION ALL the null-key rows back in unchanged; they need no dedup and must not pass through the partition.
Aggregate per account: count() over the raw week, count() over the union, and the ratio of the two. Report the ratio, not the difference, so accounts of very different sizes are comparable.
Cross-check against is_retry, which flags a resend of the same idempotency key: raw minus logical should sit close to the count of is_retry rows, and a large gap means clients are resending without a key and the dedup is under-counting duplicates.
Name the inversion explicitly: 5xx responses provoke client retries, so amplification rises exactly when reliability falls, and any engagement metric built on raw request counts will show growth during an outage.

Worked solution (25 min)

Filter to the ISO week with request_at >= :week_start AND request_at < :week_start + interval '7 days', and record the raw count per account as the baseline.
Build keyed_dedup: SELECT ... ROW_NUMBER() OVER (PARTITION BY account_id, api_key_id, idempotency_key ORDER BY request_at, request_id) AS rn FROM the week WHERE idempotency_key IS NOT NULL, then keep rn = 1.
Build unkeyed: the same week's rows WHERE idempotency_key IS NULL, taken as-is.
UNION ALL the two, then GROUP BY account_id selecting the logical count; join back to the raw counts and compute raw::numeric / logical.
Validate on one account by comparing raw - logical against COUNT(*) FILTER (WHERE is_retry) for that account and explaining any gap.

Expected result: One row per account with raw_requests, logical_operations and an amplification ratio at or above 1.0, where logical_operations counts each distinct (account_id, api_key_id, idempotency_key) once and each null-key request once.

Verify it yourself

logical_operations is never greater than raw_requests, and the two are equal for accounts whose keys are all distinct.
For accounts with no null idempotency_key, raw_requests - logical_operations is close to the count of is_retry rows.
Deliberately grouping on the key without the null split reproduces a visibly lower count, confirming the NULL collapse is the mechanism.
Re-running the dedup returns the same surviving request_id set, proving the ORDER BY tiebreak made it deterministic.

Follow-up

Which of these two counts belongs in the billable-units metric, and which in an engagement metric?
An account's amplification ratio jumps from 1.05 to 3.0 in a day. Name three causes and the query that separates them.

What metrics would you track for a product launch involving a new feature?

Medium · product sense and metrics

Approach

Decompose the metric into the rates that drive it, and say which one you would check first.
Name one primary metric, then the guardrail that stops it being gamed.
Fix the population and the time window before naming any metric.

Follow-up

How would you detect that the metric is being gamed rather than genuinely improving?
Which segment would you cut first, and what would that rule out?

How would you approach a project to predict user churn for an application?

Medium · product sense and metrics

Approach

Decompose the metric into the rates that drive it, and say which one you would check first.
Fix the population and the time window before naming any metric.
Name one primary metric, then the guardrail that stops it being gamed.

Follow-up

Which segment would you cut first, and what would that rule out?
What would you do if the primary metric and the guardrail moved in opposite directions?

If you were tasked with improving a recommendation system, what data would you prioritize?

Medium · product sense and metrics

Approach

Fix the population and the time window before naming any metric.
Decompose the metric into the rates that drive it, and say which one you would check first.
State what result would change your recommendation, so the answer is falsifiable.

Follow-up

What would you do if the primary metric and the guardrail moved in opposite directions?
Which segment would you cut first, and what would that rule out?

How do you prioritize tasks when working on multiple projects?

Medium · product sense and metrics

Approach

Fix the population and the time window before naming any metric.
Name one primary metric, then the guardrail that stops it being gamed.
State what result would change your recommendation, so the answer is falsifiable.

Follow-up

How would you detect that the metric is being gamed rather than genuinely improving?
Which segment would you cut first, and what would that rule out?

Describe how you would validate the results of an A/B test.

Medium · a/b testing and experimentation

Approach

Name the randomisation unit first; it decides the variance and what the test can detect.
Name the guardrails that would stop a launch even on a positive primary result.
Say whether units interfere with each other, and switch design if they do.

Follow-up

What would you do if you could not randomise at all?
How would you handle interference between treated and control units?

Explain the difference between supervised and unsupervised learning.

Medium

Approach

Say what you would check first and why it is the highest-information step.
State your assumptions explicitly before working the problem.
Work from the decision backwards to the evidence you would need.

Follow-up

How would you know your answer was wrong?
What assumption would you test first?

Estimate a staggered overage-price change without a control arm

Hard · difference-in-differences · synthetic-control · staggered-adoption

On 2026-03-01 the overage rate for sku_code = 'ingest_gb' rose 18% for accounts whose dim_account.billing_country falls in six countries. Each account moves to the new rate at its own renewal, so treatment switches on across twelve months. Randomisation was never available. Using fct_usage_daily and fct_subscription_period, estimate the effect on net_amount_cents per account and on gross logo retention. Specify the estimator, the identifying assumption and how you test it, how you handle accounts that churn out of the panel, and how you do inference with six treated countries.

Approach

Fix the estimand before the estimator. Effect on revenue per account and effect on retention are different questions, and the second determines whether the first is worth having: an 18% rate rise that raises revenue while costing renewals is a loss. Define both on a cohort frozen before 2026-03-01, carrying churned accounts at zero revenue rather than dropping them, so attrition cannot masquerade as a revenue effect.
Do not run a single two-way fixed-effects regression. With treatment switching on at different renewal dates and effects that vary with time since treatment, the two-way fixed-effects coefficient is a variance-weighted average that includes comparisons of later-treated accounts against already-treated ones, and those comparisons can carry negative weights. Use a cohort-by-cohort estimator such as Callaway-Sant'Anna, or a stacked event study using only not-yet-treated controls, and report the event-study path rather than one number.
Treat parallel trends as a maintained assumption to be probed, not a result to be shown. Plot pre-treatment lead coefficients with intervals and state that flat leads are consistent with the assumption rather than proof of it. Handle anticipation explicitly: accounts notified of the increase before their renewal can pull ingest volume forward, so exclude a pre-renewal anticipation window and check whether the leads move when you do.
When the six treated countries are dominated by a few large accounts, the group mean is not a stable object and difference-in-differences on it will be noisy in a way its standard error will not reflect. Build a synthetic control at the country level from a donor pool of untreated countries, weighted to match the pre-period trajectory of net revenue per account, and report the placebo distribution across donor countries instead of a conventional standard error.
Do inference at the level treatment was assigned, which is the country. Six treated clusters is far too few for cluster-robust standard errors, which are badly downward-biased below roughly forty clusters. Use a wild cluster bootstrap, and report a randomisation-inference p-value from placebo assignments of treated status alongside it.
Apply the domain's data hygiene or the whole estimate is contaminated: exclude is_internal accounts, exclude the trailing metering settling window so the final months are not artificially low, and join as-of to the fct_subscription_period version live on each date rather than to is_current, which would price last year's usage at this year's contract.

Worked solution (45 min)

Build a balanced monthly account panel from fct_usage_daily joined as-of to fct_subscription_period, excluding is_internal accounts and carrying churned accounts forward at zero revenue.
Define treatment cohorts by renewal month, estimate group-time average treatment effects, and aggregate to an event-study path with leads from -6 to -1 and lags from 0 to +11.
Fit a country-level synthetic control on pre-period net revenue per account and generate the placebo distribution by refitting on each donor country in turn.
Compute a wild cluster bootstrap p-value clustered on country and report the randomisation-inference p-value from the placebo distribution beside it.
Repeat the full path for gross logo retention, restricted each month to the renewal-eligible base plus the 45-day grace window.

Expected result: An event-study path whose lead coefficients are statistically indistinguishable from zero and whose lags are reported by months since renewal, with country-level inference from a wild cluster bootstrap and a placebo-based p-value from the synthetic control. A single pooled two-way fixed-effects number is not reported, and the retention path is computed only on the renewal-eligible base.

Verify it yourself

Treated and synthetic pre-period trajectories must track within the pre-period fit error. A poor pre-period fit cannot be rescued by a longer post-period and invalidates the comparison.
Gross logo retention must be computed on accounts whose term_end_date fell in the month, not on the whole panel. On annual contracts only about a twelfth of accounts can churn in any month, so the whole-base rate understates churn by roughly an order of magnitude.
At six treated clusters the wild cluster bootstrap p-value should be noticeably larger than the naive cluster-robust one. If they agree closely, the bootstrap was run on the wrong clustering variable.

Follow-up

Your event study shows a significant lead coefficient two months before renewal. Is that anticipation, a violation of parallel trends, or a coding error, and what distinguishes them?
Revenue per account rose 6% and gross logo retention fell 1.8 points. How do you combine those into a single recommendation, and over what horizon?
One treated country contains an account holding 20% of that country's revenue. What does that do to the synthetic control fit, and what would you do about it?

Weekly active organisations fell nine percent over one week

Easy · seasonality · denominators · segmentation

A dashboard reports the weekly active organisation ratio on a trailing seven-day window ending each Wednesday. This week it reads nine percent below last week. You have fct_api_request (account_id, environment, traffic_class, http_status, request_at) and dim_account (account_id, billing_country, account_status, is_internal, is_current). Nothing was released. Decide whether usage actually fell, and hand back a corrected series plus a one-paragraph explanation that a non-analyst can repeat without you in the room.

Approach

Count the holiday-free business days inside each window before comparing them. A trailing seven-day window spans exactly five weekdays wherever it ends, so its business-day count can only fall to four or fewer when a public holiday lands inside it and can never reach six, while usage in this domain follows a hard five-to-two weekday cycle. Two windows holding different numbers of business days are not comparable whatever the product did.
Count distinct active accounts per calendar day for the last ten weeks and overlay the two windows. A calendar problem shows as a small number of weekdays sitting at weekend level, not as every day being uniformly lower.
Cut the daily series by billing_country and index each country-day to that country's trailing same-weekday median, which isolates a regional public holiday from a product change.
Check the denominator on its own: the metric divides by accounts whose account_status was in trial, free or active_paid for the whole week, so a batch suspension or status backfill moves the ratio with no change in the numerator at all.
Report the series with each window's business-day count and the holiday dates annotated beside it, and state the residual week-over-week change that survives once the calendar effect is removed. Compare against earlier windows holding the same number of business days rather than dividing by business days, since distinct-account counts are sublinear in window length and dividing would over-correct.

Follow-up

Distinct account counts are sublinear in the number of days in the window. Why does losing one of five business days reduce the count by noticeably less than twenty percent?
How would you make this metric comparable across countries with different holiday calendars without hand-maintaining a holiday table forever?

Suggested practice checklist

Four days spend equal time on query work, statistics, modelling and product judgement at deliberately shallow depth, which produces a scored map of where you actually stand. The last three days spend everything on the two areas the role weights most, and close by re-running day one to measure movement.

Day 1 — Breadth pass: query fluency

Solve six prompts spanning aggregation, joins, window functions and date arithmetic in 60 minutes total, stopping at 10 minutes each whether or not it works, and mark every prompt as solved, solved slowly, or stuck.
For each unsolved prompt write the single blocking sentence (I lost the grain, I did not know the frame clause, I could not express the date boundary) instead of reading the solution.
Translate one pandas transformation you know well into SQL and one SQL query into pandas, checking that both return the same row count and the same totals.

Deliverable: A scored six-row table, one line per prompt, saved for the day-seven re-run.

Day 2 — Breadth pass: statistics and inference

Answer ten short questions in writing with nothing open: what a p-value is conditional on, what a 95 percent interval covers across repeated samples, when a paired test is the right one, what the bootstrap estimates, why multiple comparisons inflate false positives, how controlling the family-wise error rate differs from controlling the false discovery rate, what power depends on, what a missed real effect costs a product, the three situations where the central limit theorem does not rescue you (small n, very heavy tails, dependent observations), and what a standard error is the standard deviation of.
Grade yourself against a reference and count only the answers that were exactly right, not the ones that were nearly right.
Rewrite the two weakest answers the following morning from memory in full sentences.

Deliverable: Ten graded answers with an honest count of exact hits.

Day 3 — Breadth pass: modelling

Take one tabular dataset end to end in 90 minutes: a leakage-safe split, a baseline that is not a model (majority class or historical mean), one regularized linear model, one gradient-boosted tree, and a single evaluation metric chosen before you look at any result.
Write why that metric fits the cost structure: precision at a fixed recall for alerting, calibration for anything feeding a price or a threshold, ranking metrics for retrieval, and note that area under the ROC curve is insensitive to class balance in a way that can flatter a rare-positive problem.
Name the leak you were most likely to introduce (an encoding fit on all rows before splitting, or a feature computed after the label's timestamp) and write the check that would have caught it.

Deliverable: A notebook whose first cell states the metric and the baseline, plus two lines on what beat what and by how much.

Day 4 — Breadth pass: product judgement

Answer three case prompts aloud at 15 minutes each, timing how long passes before you state a success metric.
For one case write the first segmentation you would run and the row counts you expect per segment, so that a tiny segment cannot quietly drive the conclusion.
Take a metric definition you did not write, from a public dashboard, a textbook, or documentation you already have open, and list every place two analysts implementing it would diverge: which rows the denominator admits, whether the unit is an account or a person, what the time window is anchored to, and what happens to data that arrives late. Then write the one question that would close the largest of those gaps.

Deliverable: Three recorded case answers plus an ambiguity list for a metric someone else defined, ending in the single question you would ask about it.

Day 5 — Depth, first area

Rank the four areas by how many bullet points in the role description each one covers, pick the top one, and spend the entire day inside it.
Work the six hardest problems you can find in that area and for each write the generalizable move you should have reached for first, rather than the answer.
Re-solve the two you failed the same evening with notes closed.

Deliverable: Six generalizable moves written as instructions to yourself, not as solutions.

Day 6 — Depth, second area, and the seam between them

Repeat the depth protocol on the second-ranked area with the same six-problem structure.
Construct one problem that requires both areas at once, for example a metric redefinition whose effect you must validate with a test whose readout you then have to query.
Solve your own combined problem end to end and note where the handoff between the two areas cost you time.

Deliverable: One combined problem, solved end to end, with the handoff failure written down.

Day 7 — Integration and re-measurement

Re-run the six prompts from day one under the same clock and compare both correctness and time.
Run a 60-minute mixed mock that moves between areas without warning, since switching cost is what breadth passes do not train.
Write the two areas you would still fail on, and the sentence you will use in the interview when you hit one of them.

Deliverable: A before-and-after score table plus a written plan for the two remaining gaps.

Bring numbers, including the ones you are unsure of

An answer without a quantity is hard to interrogate, so interviewers keep probing until they find one. Come with the baseline, the change, the window it was measured over, and how confident you were. If the effect never got measured, say so and say what you would have measured. Fabricated precision is worse than an honest gap.

Describe a situation where you demonstrated a commitment to diversity and inclusion.

Medium · behavioural and stakeholder questions

Approach

Close with what you would do differently, concretely.
Pick a story where you drove the decision, not one where you observed it.
State the situation in two sentences and spend the rest on your reasoning.

Follow-up

How did you know the outcome was caused by your change?
What did you decide not to do, and why?

Walk through an analysis you later discovered was wrong

Easy · data quality · error ownership · metering

Six weeks ago you reported that consumption fell 9 percent in the last week of the month, and a team spent a sprint investigating the cause. The fall was an artefact: rows in fct_usage_daily land late and are restated in place, and you queried before the tail had settled. Describe how you found the error, what you told the people who acted on it, and the control you put in place so this class of mistake cannot reach a dashboard again. Be specific about how the settling window was measured.

Approach

The interviewer is probing whether you self-report errors before someone else finds them, and whether your fix is structural rather than a promise to be more careful. Say plainly that the number was wrong and that a sprint was spent on it, before describing any diagnosis.
Establish the artefact quantitatively instead of asserting that data lands late. For each usage_date, compare the total as of first_written_at against the settled total and read the settling time off that curve, for example 97 percent of final by day three and 99.5 percent by day five.
Correct the record the same day, in the channel the original number went out in, to the same audience. The cost of the wasted sprint belongs in the correction, not in a footnote.
Make the fix structural: exclude a trailing lag window from every reportable figure, and make the reporting view return no rows inside that window rather than returning partial ones. A dashboard that shades unsettled days still gets read as a decline.
State what generalises. Any fact table restated in place has this failure mode, so the guard belongs at the source rather than on the one dashboard that embarrassed you. A strong answer ends with the class of error closed; a generic one ends with a lesson learned.

Follow-up

How did you choose the completeness threshold behind the lag window, and what would make you recalibrate it?
What did you say to the team that lost the sprint, and what did they say back?
Is there a legitimate case for showing the unsettled tail at all, and to whom?

Defend a churn number twelve times the one in the board deck

Hard · stakeholder management · retention metrics · definitions

You recompute logo churn on the renewal-eligible base from fct_subscription_period, counting only accounts whose term_end_date fell in the month and allowing a 45-day grace for late paperwork. Annualised, about 14 percent of accounts that reach a renewal date do not renew. A revenue leader has been quoting 1.2 percent to the board for two quarters, computed by dividing non-renewed accounts by the entire customer base in each month and printing that monthly figure with no period attached. You have 20 minutes with that leader and the finance lead. Decide which number is reported from now on, and what happens to the two quarters already published.

Approach

The interviewer is probing whether you can hold a correct definition under social pressure without turning it into a competence dispute. Open by reproducing their 1.2 percent exactly, with their denominator and their months, so the disagreement is arithmetic both sides can see rather than a claim about who was careless.
Separate the two defects, because they are different in kind. The denominator is wrong: on annual contracts only about one twelfth of the base reaches a renewal date in any month, so an account eleven months from renewal sits in the denominator while being structurally incapable of entering the numerator, which suppresses the rate by a factor near twelve. The period is merely unstated: a monthly figure printed beside annual revenue targets gets read as an annual rate.
Say out loud that those two defects nearly cancel in the level, before the leader finds it. Twelve times 1.2 percent is about 14 percent, which is your number. That is the strongest thing you can say in the room, because it proves both figures rest on the same non-renewal count and moves the meeting onto which denominator and which period get published rather than onto whose query is right.
The level is recoverable; the series is not. Non-renewals in a month are the eligible base for that month times the churn rate, so dividing by a fixed whole base makes the published line proportional to how many contracts happen to come up that month. Where signings cluster at quarter ends, the eligible base in a quarter-end month can be several times a quiet month's, and the month-over-month moves the board has been reading as satisfaction are the signing calendar.
Separate the measurement change from a business change. Nothing got worse this week; the loss rate was always this. Bring net revenue retention over the same period as a ratio of sums on a cohort frozen twelve months earlier, because logo churn concentrated in small accounts can sit beside healthy revenue retention, and that combination is the actual story.
Offer a migration path rather than a correction. Report both rates for one quarter with a written bridge, restate the prior two quarters in an appendix instead of silently, and pin the definition, including the period it is stated over, somewhere finance and product both read it. Concede the limits of your own number: the 45-day grace means the most recent 45 days are not reportable, and churn must be dated on term_end_date rather than on updated_at. A strong answer volunteers this; a generic one only defends.

Follow-up

The leader multiplies their monthly figure by twelve, lands on your annual number, and concludes nothing was ever wrong. What do you say?
The leader says publishing the corrected rate costs the team its credibility with the board this quarter. What do you do?
Gross logo retention worsened while net revenue retention improved. Which do you lead with, and what does the combination tell you about who is leaving?