AI Research Institute Data Scientist Interview Questions & Guide 2026

The 60-second brief

As a Data Scientist at AI Research Institute, you serve as the analytical engine behind the institute's mission to push the boundaries of machine learning and artificial intelligence. Your work is not merely theoretical; you are responsible for translating complex data patterns into actionable insights that inform the institute's core research initiatives and product development. By bridging the gap between raw data and high-level strategy, you help the institute's teams make evidence-based decisions about intelligent systems.

Allocate prep to your weakest link rather than your favourite topic. Of the three things that usually gate the outcome (SQL that is correct under messy joins, sound reasoning about experiments, and structured framing of an open-ended problem), candidates tend to over-invest in modelling theory and under-invest in framing.

Focus areas

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

Overview

The role is highly dynamic, requiring a balance of technical rigor and business acumen. You will navigate large, often unstructured datasets, design experiments to validate research hypotheses, and communicate findings to cross-functional stakeholders. Success in this position means you are comfortable with ambiguity, capable of building robust analytical frameworks, and eager to contribute to a collaborative, fast-paced environment where innovation is the primary metric of success.

Where to focus your preparation

Treat the sections below as preparation areas and confirm the format with your recruiter.

Preparation focus

No round sequence has been reported for this company, so work the categories below and confirm the format with your recruiter.

What to demonstrate

• Breadth across SQL, experimentation and product reasoning
• Ability to state assumptions before choosing a method

How to prepare

• Drill the practice exercises below and time yourself
• Prepare three quantified stories about decisions you drove

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.

Comparing accounts that received a sales or customer-success touch against those that did not

Assignment of coverage is deliberate and pulls in both directions at once: the largest accounts get a named owner because they are valuable, and the accounts showing distress get one because they are at risk. The comparison therefore mixes a strong positive selection with a strong negative one, and the naive estimate can come out with either sign depending on which assignment rule dominated during the period examined. Nothing about matching on observed size fixes this, because the risk signal that triggered coverage is usually the same signal that predicts the outcome. It needs either an actual randomised or staggered rollout of coverage, or a design built on a capacity constraint or territory boundary that assigns coverage for reasons unrelated to account health.

Analysing at a different unit than the one randomised

Say out loud what was randomised (user, device, account, cluster) and make the analysis unit match, or account for the clustering with cluster-robust standard errors, the delta method, or aggregation up to the randomised unit. Randomising users and then running a test over sessions understates variance and inflates the false-positive rate.

Never asking what decision the analysis will inform

Open with who makes the decision, what the options are, and by when. The answer determines the precision you need, the segments worth cutting, and whether an observational read suffices or an experiment is required.

Questions & practice

How would you design a test to measure the efficacy of a new algorithm?

Medium · machine learning and modelling

Approach

1. Frame the prediction: the label, the moment of prediction, and the action it triggers.
2. Say how the offline result would be validated online before it is trusted.
3. Set a baseline first, so any model has something honest to beat.

Follow-up

• Where could label leakage enter this setup?
• What would you monitor after launch to know the model is still valid?

Sessionise an API event stream with a 30-minute inactivity gap

Medium · sessionisation · event-streams · pandas

fct_api_request arrives as a DataFrame with account_id, user_id, request_at (tz-aware UTC), traffic_class and http_status, roughly 5 million rows. Assign a session_id to every human-attributable request: drop rows where user_id is null or traffic_class is in ('ci','synthetic_monitor','load_test'), then open a new session whenever the gap since that user's previous remaining request exceeds 30 minutes. Return the filtered frame plus session_id, and a per-session summary with user_id, account_id, session start, session end and request count. Do not loop over rows.

Approach

1. Settle the filter-then-gap ordering before writing code. Removing CI and synthetic rows changes the gaps, so sessionising the raw stream and filtering afterwards is a different answer; the definition given filters first, and the two diverge most for accounts whose CI runs every ten minutes.
2. Sort once by (user_id, request_at) with a stable kind, then gap = df.groupby('user_id', sort=False).request_at.diff(). The first row of each user yields NaT, which is exactly the boundary condition you want rather than a special case to patch.
3. new_session = gap.isna() | (gap > Timedelta(minutes=30)); session_id = new_session.cumsum(). The cumsum runs over the whole sorted frame and therefore produces globally unique ids in one pass; a per-user cumcount collides across users and forces a composite key on every downstream join.
4. Build the summary with a single groupby('session_id').agg(...). user_id and account_id can be carried with 'first' only because the sort key groups them — state that dependency, since it silently breaks if someone later re-sorts the frame.
5. Decide explicitly what a session means when one user_id holds memberships in several accounts: either add account_id to the sort and group keys, or document that sessions may cross accounts. Leaving it undecided produces sessions whose account_id is whichever row sorted first.

Worked solution (30 min)

1. human = df[df.user_id.notna() & ~df.traffic_class.isin(['ci','synthetic_monitor','load_test'])].sort_values(['user_id','request_at'], kind='mergesort').reset_index(drop=True)
2. gap = human.groupby('user_id', sort=False).request_at.diff(); human['session_id'] = (gap.isna() | (gap > pd.Timedelta(minutes=30))).cumsum()
3. summary = human.groupby('session_id').agg(user_id=('user_id','first'), account_id=('account_id','first'), start=('request_at','min'), end=('request_at','max'), n_requests=('request_at','size')).reset_index()
4. assert summary.n_requests.sum() == len(human) and human.groupby('session_id').user_id.nunique().max() == 1

Expected result: human gains a dense session_id increasing in (user_id, request_at); summary has exactly one row per session, n_requests sums to len(human), and every within-session consecutive gap is at most 30 minutes while every same-user between-session gap exceeds it.

Verify it yourself

• summary.n_requests.sum() == len(human); no request is lost or double-counted.
• human.groupby('session_id').user_id.nunique().max() == 1.
• human.groupby('session_id').request_at.diff().max() <= pd.Timedelta(minutes=30).
• Shifting every timestamp by a constant offset leaves the session count unchanged.

Follow-up

• Where does 30 minutes come from, and how would you pick it from this data instead of from convention?
• An engineer reused their personal key for a nightly batch job, so machine traffic carries a human user_id. How would you detect that, and should those requests form sessions?
• How much does the session count change if you sessionise before dropping CI traffic rather than after?

Join daily usage to the contract version live that day

Medium · as-of-join · pandas · numpy · contracts

fct_usage_daily has account_id, usage_date and net_amount_cents. fct_subscription_period has account_id, subscription_period_id, plan_tier, term_start_date, term_end_date, arr_cents and is_current, with one row per contract term version and every amendment inserting a new row. Attach to each usage row the subscription_period_id whose term brackets usage_date (term_start_date <= usage_date <= term_end_date). pd.merge_asof and an interval-condition merge are unavailable; use sorting and numpy.searchsorted. Then report monthly net revenue by plan_tier. Terms for one account do not overlap, and usage may fall outside every term.

Approach

1. Say out loud why the cheap version is wrong: joining on is_current stamps today's plan tier onto last year's usage, so every account that upgraded has its history reclassified and revenue-by-tier becomes a function of when the query ran.
2. Sort each account's terms by term_start_date and use np.searchsorted(term_start, usage_date, side='right') - 1 to get the last term that started on or before the usage date. Do it per account group, or globally after encoding (account_id, date) into one monotone key.
3. searchsorted only enforces the left edge. Validate the right edge afterwards — usage_date <= the candidate's term_end_date — and set the match to NA where it fails. That NA is usage in a gap between contracts and must stay visible instead of being folded into the expired term.
4. Assert non-overlap before trusting the lookup, and write the assertion so it is capable of passing. prev_end = terms.groupby('account_id').term_end_date.shift() is NaT on each account's first row, and NaT < Timestamp evaluates to False rather than NA, so a comparison followed by .fillna(True) has nothing left to fill and the assertion fires on every account's first term whatever the data looks like. Guard the null yourself: assert (prev_end.isna() | (prev_end < terms.term_start_date)).all(). The failure mode of getting this wrong is not a false alarm you notice once — it is an assertion someone deletes because it never passes, after which overlapping terms make searchsorted return one of them with no trace in the output.
5. Aggregate after the join, grouping by (usage_date month, plan_tier) with dropna=False so the unmatched bucket appears as its own row and the total still ties to the ungrouped sum of net_amount_cents.

Follow-up

• An amendment takes effect on the 17th of a month. How do you report that month's revenue by tier?
• What changes if terms can overlap because of a co-term amendment?
• How would you verify this against a SQL implementation using a BETWEEN condition?

Find paid seats that never called the API

Easy · anti-join · null semantics · seat utilisation

dim_user_membership carries user_id, account_id, seat_type, is_service_account, deactivated_at and last_seen_at. fct_api_request carries user_id, which is null whenever the caller is a service account or an unattended key, plus account_id, request_at and http_status. For a single account, list every licensed_paid seat with is_service_account = false and deactivated_at null that issued no request at all in the trailing 90 days, returning user_id and last_seen_at. A first draft writes NOT IN against a subquery over fct_api_request.user_id. State exactly what that draft returns and why, then write the correct query.

Approach

1. Read the column comment first: user_id is nullable on the fact, so the subquery almost certainly contains at least one NULL for any account that runs a service account or an unattended key.
2. Work the three-valued logic out loud. x NOT IN (a, NULL) expands to NOT (x = a OR x = NULL); the second disjunct is UNKNOWN, so for any x not equal to a the whole predicate is UNKNOWN and the row is filtered out. The draft returns zero rows, which reads as full seat utilisation.
3. Rewrite as NOT EXISTS with the 90-day and status predicates inside the correlated subquery. Putting them in the outer WHERE instead turns the anti-join into a different question and silently changes the answer.
4. Correlate on both user_id and account_id. A membership is (user_id, account_id) and one person can hold memberships in several accounts, so correlating on user_id alone marks a seat as active because that human was busy somewhere else.
5. Filter the seat side to seat_type = 'licensed_paid', is_service_account = false and deactivated_at IS NULL, and say what the resulting count means next to contracted_seats on the current subscription row.

Worked solution (15 min)

1. Confirm the hazard with one query: SELECT count(*) FROM fct_api_request WHERE user_id IS NULL AND account_id = :account_id. Any non-zero result proves the draft returns nothing.
2. Write the seat side: SELECT user_id, last_seen_at FROM dim_user_membership WHERE account_id = :account_id AND seat_type = 'licensed_paid' AND NOT is_service_account AND deactivated_at IS NULL.
3. Attach AND NOT EXISTS (SELECT 1 FROM fct_api_request r WHERE r.user_id = m.user_id AND r.account_id = m.account_id AND r.request_at >= now() - interval '90 days').
4. Compare the row count against the same query written as a LEFT JOIN with a WHERE r.user_id IS NULL; the two must agree exactly.
5. Divide the active seat count by contracted_seats from the account's current fct_subscription_period row and state the utilisation figure.

Expected result: A list of user_id and last_seen_at for licensed, non-service, active memberships on that account with no request in 90 days. The NOT IN draft returns zero rows whenever the account has any service-account or unattended-key traffic, and returns the right answer only by accident on accounts that have none.

Verify it yourself

• The NOT EXISTS result and the LEFT JOIN ... IS NULL result return identical user_id sets.
• Idle seats plus seats with at least one request in 90 days equals the total licensed non-service active membership count for the account.
• Forcing a NULL into the subquery (union a SELECT NULL) leaves NOT EXISTS unchanged and empties the NOT IN version, which demonstrates the failure on demand.

Follow-up

• Adding AND user_id IS NOT NULL to the subquery also fixes NOT IN. Why is NOT EXISTS still the form you would leave in the repository?
• last_seen_at looks like a shortcut for the whole query. What does it actually record, and where does it disagree with API activity?
• This is a seat-reduction risk list. What threshold would you attach before handing it to an account team, and what happens to seats below it?

Sessionise interactive API traffic with a thirty-minute inactivity gap

Medium · window functions · sessionisation · gaps and islands

fct_api_request carries account_id, user_id (null for service accounts), request_at, traffic_class and http_status. Using only traffic_class = 'interactive' rows with a non-null user_id, group each seat's requests into sessions on a 30-minute inactivity threshold: a request more than 30 minutes after the previous request from the same (account_id, user_id) opens a new session. For one ISO week return, per account, the session count, the median session duration in minutes and the median requests per session. A single-request session has a duration of zero.

Approach

1. Filter first: traffic_class = 'interactive' and user_id IS NOT NULL. Machine traffic has no sessions in any useful sense, and leaving CI or batch rows in produces sessions that are really cron schedules.
2. Get the previous timestamp with LAG(request_at) OVER (PARTITION BY account_id, user_id ORDER BY request_at). Partitioning by user_id alone stitches one person's work across two different accounts into one fabricated session, because a human holds memberships in several accounts.
3. Flag a boundary where the lag is NULL or request_at - lag > interval '30 minutes'. Decide and state whether exactly 30 minutes continues the session; strictly greater is the conventional choice and needs to be written down either way.
4. Assign session ids with SUM(boundary::int) OVER (PARTITION BY account_id, user_id ORDER BY request_at ROWS UNBOUNDED PRECEDING), the standard running-count construction for islands.
5. Roll up to sessions with min(request_at), max(request_at) and count(*), then to accounts with percentile_cont(0.5) WITHIN GROUP (ORDER BY ...). Use medians, not means: session length is strongly right-skewed and one long-running client dominates the average.

Follow-up

• Sessions belonging to colleagues in one account are correlated. What does that do to a t-test on session length across an experiment arm?
• A long-poll or streaming endpoint keeps a connection open for hours. How do you stop it reading as one twelve-hour session?

Design a seat-utilisation metric before enforcing licensed seat limits

Medium · seat utilisation · denominator design · service accounts · renewal timing

Seat limits are currently unenforced. Leadership wants a metric to decide whether to enforce them at renewal. You have dim_user_membership (user_id, account_id, role, seat_type, is_service_account, activated_at, deactivated_at, last_seen_at) and fct_subscription_period (account_id, contracted_seats, pricing_model, term_end_date, arr_cents, is_current). Define licensed seat utilisation at the account grain, state every exclusion in both the numerator and the denominator, and say what a value above 1.0 means. Deliver the metric plus the guardrail that protects against enforcement destroying more ARR than it recovers.

Approach

1. Numerator: distinct user_id in dim_user_membership with seat_type = 'licensed_paid', is_service_account = false, deactivated_at null and last_seen_at inside a trailing 28 days. Service accounts hold a membership and burn API keys but consume no human attention, so counting them inflates utilisation hardest on the most deeply integrated accounts, which is exactly backwards.
2. Denominator: contracted_seats from the fct_subscription_period version current at the evaluation date. Exclude accounts with pricing_model = 'pure_consumption' entirely, because contracted_seats is null there and the ratio is undefined rather than zero. A null coerced to zero produces an infinite ratio that will lead the report.
3. Interpret both tails and say what each means commercially. Above 1.0 means the entitlement is not enforced and humans are sharing seats, which is recoverable revenue. Persistently below roughly 0.5 is the most reliable available signal of a seat reduction at renewal, because buyers audit seat counts at renewal and at no other moment.
4. Guardrail: ARR at risk from enforcement, the sum of arr_cents over accounts with utilisation above 1.0 whose term_end_date falls inside the enforcement window. Enforcing on an account that responds by cutting seats at renewal converts a recoverable overage into a permanent downgrade, and the recovered revenue can be smaller than the loss.
5. Report the metric as a distribution with the two tails counted separately and the renewal calendar attached, because no account can act on enforcement until its own term ends. A fleet average of seat utilisation is a number nobody can act on.

Worked solution (20 min)

1. Build the current contract per account from fct_subscription_period where is_current = true, keeping contracted_seats, pricing_model, term_end_date and arr_cents, and drop pure_consumption rows.
2. Count active humans per account from dim_user_membership under the four filters, using distinct user_id per account_id, which is safe because the table is one row per membership and not one row per human.
3. Compute the ratio and bucket accounts into below 0.5, 0.5 to 1.0, and above 1.0, joining arr_cents and term_end_date.
4. Report ARR in each bucket and the share of that ARR renewing in the next two quarters.

Expected result: Three utilisation buckets with ARR and renewal timing attached, and a metric definition whose numerator excludes service accounts, guests, free collaborators and deactivated memberships, and whose denominator is contracted_seats from the contract version current at the evaluation date, with pure-consumption accounts excluded rather than counted as zero.

Verify it yourself

• No account with pricing_model = 'pure_consumption' may appear, with a null or an infinite ratio.
• Bucket ARR must sum to the total ARR over included accounts.
• Summing per-account active-user counts across the fleet must be greater than or equal to the fleet-wide count of distinct user_id, because one person can hold memberships in several accounts; exact equality means multi-account users were deduplicated at the wrong grain.

Follow-up

• last_seen_at is null for service accounts by definition. If the is_service_account flag is unreliable, what happens to your numerator, and how would you audit the flag?
• Why a 28-day activity window rather than 7 or 90, and what evidence would change your answer?

Measure named coverage at an assignment threshold you cannot randomise

Hard · regression-discontinuity · instrumental-variables · selection-bias

Named customer-success coverage is assigned to accounts whose arr_cents at fiscal-year start is at least $50,000. About 88% of accounts above the line receive a named owner and about 9% below it do, through documented exceptions. Roughly 200 accounts sit within $15,000 of the cutoff, and gross logo retention on the renewal-eligible base runs near 0.88. Estimate the effect of coverage on renewal, name the design, state its identifying assumptions and the test for each, and say whether this data can answer the question at all.

Approach

1. Say first why the naive comparison is unusable, because the answer turns on it. Coverage is assigned to the largest accounts because they are valuable and to distressed accounts because they are at risk, so selection runs in both directions at once and the naive estimate can come out with either sign depending on which rule dominated. Matching on observed size does not fix it: the risk signal that triggered coverage is the same signal that predicts renewal.
2. Name the design and the estimand together. This is a fuzzy regression discontinuity with the $50,000 threshold as an instrument for coverage. The estimator is the Wald ratio, the jump in renewal at the cutoff divided by the jump in coverage probability of 0.88 - 0.09 = 0.79, and the estimand is a local average treatment effect for accounts near $50,000 whose coverage is actually decided by the rule, not the average effect of coverage across the book.
3. Fit it as local linear regression on each side with a triangular kernel and a data-driven bandwidth (Imbens-Kalyanaraman, or Calonico-Cattaneo-Titiunik with bias-corrected robust intervals). A global polynomial fitted across the whole ARR range is not a substitute and is well documented to manufacture spurious jumps at interior points.
4. Test the assumptions instead of asserting them. Run a density test for bunching just above $50,000, which is exactly what a sales team structuring a deal at $50,001 to win coverage would produce and which invalidates the design outright. Check continuity at the cutoff of employee_band, acquisition_channel, deployment_model and pre-period usage. Run placebo cutoffs at other ARR values, and plot the estimate against bandwidth.
5. Do the power arithmetic before promising an answer. With about 100 accounts per side and a renewal base rate of 0.88 (standard deviation 0.325), the sharp-design MDE at 80% power and two-sided 0.05 is 2.80 * 0.325 * sqrt(2/100) = about 12.9 percentage points, and dividing by the 0.79 first stage gives about 16 points on the fuzzy estimand. No plausible effect of assigning an account manager is that large.
6. Give the honest verdict with a route forward attached. This design cannot answer the question on one fiscal year. Pool several years at the same cutoff, accepting the added assumption that the coverage model and the threshold's meaning were stable across them, or ask for a randomised holdout among accounts near the threshold, which is cheap to justify precisely because the rule is already arbitrary there.

Follow-up

• The density test shows clear bunching just above $50,000. Is any part of the design salvageable, and what would you report instead?
• Coverage also correlates with support ticket severity. Does adding severity as a control help, hurt, or do nothing in a regression discontinuity?
• If you pool five fiscal years to gain power, what new identifying assumption have you taken on, and how would you check it?

Blended gross margin fell while every segment improved

Medium · mix shift · simpsons paradox · as-of join

Blended gross margin across paying accounts fell from 71 percent to 66 percent over two quarters, yet margin improved inside every plan_tier and every deployment_model. You have fct_usage_daily (account_id, sku_code, usage_date, net_amount_cents, cogs_cents), fct_subscription_period (account_id, plan_tier, term_start_date, term_end_date) and dim_account (account_id, deployment_model, is_internal, effective_from, effective_to). Produce an exact decomposition of the five-point move into mix, rate and interaction terms, and name the segment shift that carries it.

Approach

1. Write blended margin as a revenue-weighted average of segment margins, m = sum over i of w_i * m_i, where w_i is the segment's share of net_amount_cents. Margin is a ratio of sums, so only revenue weights reproduce the blended figure.
2. Apply the exact three-term decomposition: delta_m = sum(delta_w_i * m_i0) + sum(w_i0 * delta_m_i) + sum(delta_w_i * delta_m_i), read as mix, rate and interaction. It is exact by construction, so the three terms must sum to the observed change with no residual.
3. Build segments from as-of joins: take the fct_subscription_period version whose term brackets each usage_date and the dim_account version whose effective_from and effective_to bracket it, rather than joining on is_current and backdating today's attributes over last year's usage.
4. Run the decomposition on more than one segmentation, at minimum plan_tier, deployment_model and sku_code, since the mix that actually moved may not be the one anybody already suspected.
5. Once the driving segment is identified, determine whether the shift came from acquisition landing new accounts in a lower-margin segment or from existing accounts migrating, because those have different owners and different remedies.

Follow-up

• The interaction term is sometimes large. What does it mean operationally, and when would you prefer a log-mean decomposition that distributes it across the other two terms?
• Revenue-weighted average margin is one construction. What changes if leadership wants margin weighted by account count, and which question does each version answer?

Suggested practice checklist

For a candidate whose interviews will centre on A/B testing, metric movement and causal claims. Design comes before arithmetic, arithmetic before analysis, and the week ends by rehearsing the readout rather than the derivation.

Day 1 — Design one test end to end on paper

• Take a single feature change and write the full design: randomization unit, the exact point of exposure, the primary metric with its grain, guardrails, allocation, planned duration, and the decision rule committed before any data exists.
• Write why the randomization unit must sit at or above the level where treatment can spill over, and give one case where user-level randomization is still contaminated (shared accounts or devices, or two participants in the same marketplace).
• State in advance what you will do if the primary metric is flat while a secondary metric is significant.

Deliverable: A one-page test design with a decision rule written before launch.

Day 2 — Power arithmetic until it is automatic

• Compute required sample size per arm for a binary metric with the normal approximation, n is approximately 2 times (z for alpha/2 plus z for power) squared times p(1 minus p) divided by delta squared, for baselines of 2, 10 and 40 percent at a 5 percent relative lift, and note that for a fixed relative lift the requirement falls as the baseline rises because delta grows proportionally with p.
• Redo the calculation for a continuous metric using variance in place of p(1 minus p), and show why a heavy-tailed quantity such as revenue per user needs either far more traffic or a capped version with a stated cap.
• Convert one of the results into weeks given a weekly eligible traffic figure, then list the two honest ways to shorten it (accept a larger detectable effect, or reduce variance) and write why quietly lowering the power target is a decision to miss more real wins, not a speedup.

Deliverable: A small script or sheet that maps baseline, minimum detectable effect, alpha and power to sample size and weeks, cross-checked against a published calculator.

Day 3 — Variance and the unit-of-analysis problem

• Take a ratio metric whose denominator is not the randomization unit (clicks per session, randomized by user) and compute the standard error twice, once naively at session level and once by the delta method or a user-level bootstrap, then record how much the naive version understates it.
• Implement CUPED on simulated data: choose a pre-period covariate X measured before assignment, estimate theta as Cov(Y, X) divided by Var(X), and analyse Y minus theta times (X minus its mean) in place of Y. Confirm the variance of the adjusted outcome equals the raw variance multiplied by one minus the squared correlation between Y and X, so a correlation of 0.45 removes about 20 percent of the variance and not 80.
• Now run that simulation a few hundred times and confirm the adjusted effect estimate is unbiased for the same effect rather than numerically identical to the raw one. Within any single run the two differ, sometimes by a large fraction of the true effect, because the two arms' pre-period covariate means never coincide exactly in a finite sample; they agree in expectation, which is the property that matters and the one to state out loud.

Deliverable: A notebook showing the adjusted estimator with a measurably smaller variance than the raw one, plus a repeated-simulation table showing the two estimators agreeing on average while differing run by run.

Day 4 — Validity threats you can actually test for

• Run a sample ratio mismatch check as a chi-square goodness-of-fit test against the intended allocation, and write the three causes you would chase first (assignment logged before exposure, an arm-specific redirect or load failure, bot filtering applied asymmetrically).
• Simulate peeking: generate A/A data, test daily at alpha 0.05 across 14 looks, record the inflated false positive rate, then apply an alpha-spending boundary or commit to a fixed horizon and confirm the rate returns to nominal.
• Write how you would separate a novelty effect from a durable lift using the treatment effect plotted against days since first exposure, and what shape would change your recommendation.

Deliverable: One table showing the peeking false positive rate before and after correction, plus a written SRM triage list.

Day 5 — When randomization is not available

• Write the identifying assumption for difference-in-differences (parallel trends in the absence of treatment), then plot pre-period trends for two candidate control groups and justify rejecting one of them.
• Design a switchback test for a change where user-level randomization would leak across participants, choosing a time-block length against the carryover you expect and saying how you would detect carryover in the data.
• List what an interrupted time series or a synthetic control buys you and the one thing neither can rule out: an unobserved shock that coincides with the launch.

Deliverable: A one-page memo recommending a single quasi-experimental design and naming its weakest assumption explicitly.

Day 6 — The readout query

• Write the assignment-to-exposure join that returns exactly one row per unit per experiment, and handle units appearing in both arms by excluding and counting them rather than silently keeping one.
• Compute the per-arm metric, its variance and the relative lift with a confidence interval in SQL, then reproduce the identical numbers in a notebook as a cross-check.
• Add a segment breakdown and write the sentence that keeps it from being p-hacking: segments declared in advance, everything else reported as exploratory and corrected for multiplicity.

Deliverable: A single query that outputs the full readout table, matched to a notebook recomputation.

Day 7 — Present it to someone who will not read the appendix

• Give a 10-minute readout of a real or simulated experiment in the order decision, number, uncertainty, caveat.
• Have your listener ask "can we ship it" in the case where the primary is flat and a guardrail moved, and answer with a recommendation rather than a request for more data.
• Rewrite your opening line so the recommendation lands before any methodology.

Deliverable: A one-page readout whose first line is the recommendation.

Your part, not the team's

Most data work is done by groups, so an interviewer has to work out which piece was yours. An answer that runs on 'we' for several minutes gets interrupted with a question about what you personally did, and by then the answer sounds defensive even when it is true. Mark your own contribution as you go, and name the parts that belonged to someone else instead of leaving them ambiguous. Keep a few specifics back as well, like the name of the metric or who actually objected, so a probe can be answered with something you had not already said.

Can you walk us through your previous projects and the specific impact you delivered?

Medium · behavioural and stakeholder questions

Approach

1. Pick a story where you drove the decision, not one where you observed it.
2. State the situation in two sentences and spend the rest on your reasoning.
3. Name the disagreement or constraint, and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that project again?
• How did you know the outcome was caused by your change?

What motivates you to solve complex problems in an AI-driven environment?

Medium · behavioural and stakeholder questions

Approach

1. State the situation in two sentences and spend the rest on your reasoning.
2. Close with what you would do differently, concretely.
3. Name the disagreement or constraint, and how you resolved it with evidence.

Follow-up

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

Describe a time you had to pivot your research direction based on unexpected results.

Medium · behavioural and stakeholder questions

Approach

1. Name the disagreement or constraint, and how you resolved it with evidence.
2. Pick a story where you drove the decision, not one where you observed it.
3. Quantify the outcome, including what you would not claim credit for.

Follow-up

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